Daily Papers

Audio-language models compress a speech encoder's output through a Querying Transformer (Q-Former) connector before feeding it to a large language model. We identify two failures in this compression. The connector's output vectors collapse to a single direction, and different speakers produce nearly indistinguishable outputs, with paralinguistic cues such as speaker identity, gender, and prosody lost along the way. Our method, ORCA, reverses this collapse by splitting the queries into groups whose outputs are constrained to point in different directions. On SAKURA multi-hop reasoning, ORCA gains 26.4 points over an identically trained 4B baseline, reaching 75.2% (vs. 49.0% for the 8B Audio Flamingo-3). At the connector level, the same change cuts query redundancy by 12x and raises cross-speaker variance by 75x.

The effects of speaking-style variability on automatic speaker verification were investigated using the UCLA Speaker Variability database which comprises multiple speaking styles per speaker. An x-vector/PLDA (probabilistic linear discriminant analysis) system was trained with the SRE and Switchboard databases with standard augmentation techniques and evaluated with utterances from the UCLA database. The equal error rate (EER) was low when enrollment and test utterances were of the same style (e.g., 0.98% and 0.57% for read and conversational speech, respectively), but it increased substantially when styles were mismatched between enrollment and test utterances. For instance, when enrolled with conversation utterances, the EER increased to 3.03%, 2.96% and 22.12% when tested on read, narrative, and pet-directed speech, respectively. To reduce the effect of style mismatch, we propose an entropy-based variable frame rate technique to artificially generate style-normalized representations for PLDA adaptation. The proposed system significantly improved performance. In the aforementioned conditions, the EERs improved to 2.69% (conversation -- read), 2.27% (conversation -- narrative), and 18.75% (pet-directed -- read). Overall, the proposed technique performed comparably to multi-style PLDA adaptation without the need for training data in different speaking styles per speaker.

We present a speaker-aware approach for simulating multi-speaker conversations that captures temporal consistency and realistic turn-taking dynamics. Prior work typically models aggregate conversational statistics under an independence assumption across speakers and turns. In contrast, our method uses speaker-specific deviation distributions enforcing intra-speaker temporal consistency, while a Markov chain governs turn-taking and a fixed room impulse response preserves spatial realism. We also unify pauses and overlaps into a single gap distribution, modeled with kernel density estimation for smooth continuity. Evaluation on Switchboard using intrinsic metrics - global gap statistics, correlations between consecutive gaps, copula-based higher-order dependencies, turn-taking entropy, and gap survival functions - shows that speaker-aware simulation better aligns with real conversational patterns than the baseline method, capturing fine-grained temporal dependencies and realistic speaker alternation, while revealing open challenges in modeling long-range conversational structure.

Evaluation benchmarks are the cornerstone of measuring capabilities of large language models (LLMs), as well as driving progress in said capabilities. Originally designed to make claims about capabilities (or lack thereof) in fully pretrained models, evaluation benchmarks are now also extensively used to decide between various training choices. Despite this widespread usage, we rarely quantify the variance in our evaluation benchmarks, which dictates whether differences in performance are meaningful. Here, we define and measure a range of metrics geared towards measuring variance in evaluation benchmarks, including seed variance across initialisations, and monotonicity during training. By studying a large number of models -- both openly available and pretrained from scratch -- we provide empirical estimates for a variety of variance metrics, with considerations and recommendations for practitioners. We also evaluate the utility and tradeoffs of continuous versus discrete performance measures and explore options for better understanding and reducing this variance. We find that simple changes, such as framing choice tasks (like MMLU) as completion tasks, can often reduce variance for smaller scale (sim7B) models, while more involved methods inspired from human testing literature (such as item analysis and item response theory) struggle to meaningfully reduce variance. Overall, our work provides insights into variance in evaluation benchmarks, suggests LM-specific techniques to reduce variance, and more generally encourages practitioners to carefully factor in variance when comparing models.

The performance of speaker verification systems degrades significantly under language mismatch, a critical challenge exacerbated by the field's reliance on English-centric data. To address this, we propose the TidyVoice Challenge for cross-lingual speaker verification. The challenge leverages the TidyVoiceX dataset from the novel TidyVoice benchmark, a large-scale, multilingual corpus derived from Mozilla Common Voice, and specifically curated to isolate the effect of language switching across approximately 40 languages. Participants will be tasked with building systems robust to this mismatch, with performance primarily evaluated using the Equal Error Rate on cross-language trials. By providing standardized data, open-source baselines, and a rigorous evaluation protocol, this challenge aims to drive research towards fairer, more inclusive, and language-independent speaker recognition technologies, directly aligning with the Interspeech 2026 theme, "Speaking Together."

Diffusion models trained on different, non-overlapping subsets of a dataset often produce strikingly similar outputs when given the same noise seed. We trace this consistency to a simple linear effect: the shared Gaussian statistics across splits already predict much of the generated images. To formalize this, we develop a random matrix theory (RMT) framework that quantifies how finite datasets shape the expectation and variance of the learned denoiser and sampling map in the linear setting. For expectations, sampling variability acts as a renormalization of the noise level through a self-consistent relation σ^2 mapsto κ(σ^2), explaining why limited data overshrink low-variance directions and pull samples toward the dataset mean. For fluctuations, our variance formulas reveal three key factors behind cross-split disagreement: anisotropy across eigenmodes, inhomogeneity across inputs, and overall scaling with dataset size. Extending deterministic-equivalence tools to fractional matrix powers further allows us to analyze entire sampling trajectories. The theory sharply predicts the behavior of linear diffusion models, and we validate its predictions on UNet and DiT architectures in their non-memorization regime, identifying where and how samples deviates across training data split. This provides a principled baseline for reproducibility in diffusion training, linking spectral properties of data to the stability of generative outputs.

Large-scale Speech Language Models (SpeechLMs) have enabled voice assistants capable of understanding natural spoken queries and performing complex tasks. However, existing speech benchmarks primarily focus on isolated capabilities such as transcription, or question-answering, and do not systematically evaluate agentic scenarios encompassing multilingual and cultural understanding, as well as adversarial robustness. To address this, we introduce VoiceAgentBench, a comprehensive benchmark designed to evaluate SpeechLMs in realistic spoken agentic settings. It comprises over 5,500 synthetic spoken queries, including dialogues grounded in Indian context, covering single-tool invocations, multi-tool workflows, multi-turn interactions, and safety evaluations. The benchmark supports English, Hindi, and 5 other Indian languages, reflecting real-world linguistic and cultural diversity. We simulate speaker variability using a novel sampling algorithm that selects audios for TTS voice conversion based on its speaker embeddings, maximizing acoustic and speaker diversity. Our evaluation measures tool selection accuracy, structural consistency, and the correctness of tool invocations, including adversarial robustness. Our experiments reveal significant gaps in contextual tool orchestration tasks, Indic generalization, and adversarial robustness, exposing critical limitations of current SpeechLMs.

This paper proposes RefXVC, a method for cross-lingual voice conversion (XVC) that leverages reference information to improve conversion performance. Previous XVC works generally take an average speaker embedding to condition the speaker identity, which does not account for the changing timbre of speech that occurs with different pronunciations. To address this, our method uses both global and local speaker embeddings to capture the timbre changes during speech conversion. Additionally, we observed a connection between timbre and pronunciation in different languages and utilized this by incorporating a timbre encoder and a pronunciation matching network into our model. Furthermore, we found that the variation in tones is not adequately reflected in a sentence, and therefore, we used multiple references to better capture the range of a speaker's voice. The proposed method outperformed existing systems in terms of both speech quality and speaker similarity, highlighting the effectiveness of leveraging reference information in cross-lingual voice conversion. The converted speech samples can be found on the website: http://refxvc.dn3point.com

Research on human spoken language has shown that speech plays an important role in identifying speaker personality traits. In this work, we propose an approach for identifying speaker personality traits using overlap dynamics in multiparty spoken dialogues. We first define a set of novel features representing the overlap dynamics of each speaker. We then investigate the impact of speaker personality traits on these features using ANOVA tests. We find that features of overlap dynamics significantly vary for speakers with different levels of both Extraversion and Conscientiousness. Finally, we find that classifiers using only overlap dynamics features outperform random guessing in identifying Extraversion and Agreeableness, and that the improvements are statistically significant.

Speaker embeddings aggregate frame-level acoustic features into compact representations for speaker recognition. Recent uncertainty-aware speaker modeling approaches further characterize the reliability of speaker embeddings by estimating their associated uncertainty. However, existing methods often suffer from inaccurate uncertainty estimation and uncertainty miscalibration under domain shifts. To address these challenges, we propose a robust uncertainty modeling framework from both estimation and adaptation perspectives. Specifically, we introduce an Inter- and Intra-Speaker-Aware Uncertainty Softmax that incorporates both inter-speaker separability and intra-speaker variability into uncertainty learning, enabling uncertainty estimates to better capture the reliability of speaker embeddings. Furthermore, we propose an Uncertainty-Calibrated Domain Adaptation (UCDA) framework to mitigate uncertainty miscalibration caused by domain mismatch. Extensive experiments on both in-domain and cross-domain benchmarks demonstrate that the proposed approach consistently improves uncertainty reliability and speaker recognition robustness.

A speaker encoder used in multilingual voice cloning should treat the same speaker identically regardless of which script the audio was uttered in. Off-the-shelf encoders do not, and the failure is accent-conditional. On a 1043-pair Western-accented voice corpus across English, Hindi, Telugu, and Tamil, WavLM-base-plus-sv loses 0.082 absolute cosine similarity when the same voice changes script and ECAPA-TDNN loses 0.105. On a 1369-pair Indian-accented voice corpus, the gap shrinks to 0.006 (WavLM-SV) and 0.044 (ECAPA-TDNN). The leak is largest where it matters most for cross-script TTS: when a system projects a non-Indic-trained voice into Indic scripts. We present LASE (Language-Adversarial Speaker Encoder), a small projection head over frozen WavLM-base-plus trained with two losses: a supervised contrastive loss over voice identity, and a gradient-reversal cross-entropy against a 4-language classifier that pushes the embedding to be language-uninformative while remaining speaker-informative. Trained on 1118 quality-gated cross-script pairs synthesised from 8 commercial multilingual voices, LASE's residual gap is consistent with zero on both corpora (Delta = 0.013 Western, Delta = 0.026 Indian; both bootstrap 95% CIs include zero) and amplifies the cross-script-vs-floor margin 2.4-2.7x over both baselines. An ECAPA+GRL ablation shows the GRL objective improves either backbone but the WavLM choice contributes too. In synthetic multi-speaker diarisation, LASE matches ECAPA-TDNN on cross-script speaker recall (0.788 vs 0.789) with ~100x less training data. We release the r1 checkpoint, both corpora, and the bootstrap recipe.

The development of robust, multilingual speaker recognition systems is hindered by a lack of large-scale, publicly available and multilingual datasets, particularly for the read-speech style crucial for applications like anti-spoofing. To address this gap, we introduce the TidyVoice dataset derived from the Mozilla Common Voice corpus after mitigating its inherent speaker heterogeneity within the provided client IDs. TidyVoice currently contains training and test data from over 212,000 monolingual speakers (Tidy-M) and around 4,500 multilingual speakers (Tidy-X) from which we derive two distinct conditions. The Tidy-M condition contains target and non-target trials from monolingual speakers across 81 languages. The Tidy-X condition contains target and non-target trials from multilingual speakers in both same- and cross-language trials. We employ two architectures of ResNet models, achieving a 0.35% EER by fine-tuning on our comprehensive Tidy-M partition. Moreover, we show that this fine-tuning enhances the model's generalization, improving performance on unseen conversational interview data from the CANDOR corpus. The complete dataset, evaluation trials, and our models are publicly released to provide a new resource for the community.

Large language models achieve high performance on many but not all downstream tasks. The interaction between pretraining data and task data is commonly assumed to determine this variance: a task with data that is more similar to a model's pretraining data is assumed to be easier for that model. We test whether distributional and example-specific similarity measures (embedding-, token- and model-based) correlate with language model performance through a large-scale comparison of the Pile and C4 pretraining datasets with downstream benchmarks. Similarity correlates with performance for multilingual datasets, but in other benchmarks, we surprisingly find that similarity metrics are not correlated with accuracy or even each other. This suggests that the relationship between pretraining data and downstream tasks is more complex than often assumed.

Large language models (LLMs) exhibit cognitive biases -- systematic tendencies of irrational decision-making, similar to those seen in humans. Prior work has found that these biases vary across models and can be amplified by instruction tuning. However, it remains unclear if these differences in biases stem from pretraining, finetuning, or even random noise due to training stochasticity. We propose a two-step causal experimental approach to disentangle these factors. First, we finetune models multiple times using different random seeds to study how training randomness affects over 30 cognitive biases. Second, we introduce cross-tuning -- swapping instruction datasets between models to isolate bias sources. This swap uses datasets that led to different bias patterns, directly testing whether biases are dataset-dependent. Our findings reveal that while training randomness introduces some variability, biases are mainly shaped by pretraining: models with the same pretrained backbone exhibit more similar bias patterns than those sharing only finetuning data. These insights suggest that understanding biases in finetuned models requires considering their pretraining origins beyond finetuning effects. This perspective can guide future efforts to develop principled strategies for evaluating and mitigating bias in LLMs.

Recent research in speaker recognition aims to address vulnerabilities due to variations between enrolment and test utterances, particularly in the multi-genre phenomenon where the utterances are in different speech genres. Previous resources for Vietnamese speaker recognition are either limited in size or do not focus on genre diversity, leaving studies in multi-genre effects unexplored. This paper introduces VoxVietnam, the first multi-genre dataset for Vietnamese speaker recognition with over 187,000 utterances from 1,406 speakers and an automated pipeline to construct a dataset on a large scale from public sources. Our experiments show the challenges posed by the multi-genre phenomenon to models trained on a single-genre dataset, and demonstrate a significant increase in performance upon incorporating the VoxVietnam into the training process. Our experiments are conducted to study the challenges of the multi-genre phenomenon in speaker recognition and the performance gain when the proposed dataset is used for multi-genre training.

Speaker recognition models face challenges in multi-lingual settings due to the entanglement of linguistic information within speaker embeddings. The overlap between vocal traits such as accent, vocal anatomy, and a language's phonetic structure complicates separating linguistic and speaker information. Disentangling these components can significantly improve speaker recognition accuracy. To this end, we propose a novel disentanglement learning strategy that integrates joint learning through prefix-tuned cross-attention. This approach is particularly effective when speakers switch between languages. Experimental results show the model generalizes across monolingual and multi-lingual settings, including unseen languages. Notably, the proposed model improves the equal error rate across multiple datasets, highlighting its ability to separate language information from speaker embeddings and enhance recognition in diverse linguistic conditions.

Significant disparities exist in the diagnosis and clinical presentation of depression across different linguistic populations. Speech-based depression detection performs well monolingually, but cross-lingual generalization remains an open challenge. A key reason is that prior work uses segment-level random splits without speaker grouping, leading to identity leakage that inflates reported metrics. We propose CLeaD, a supervised contrastive alignment framework that maps WavLM embeddings from English and Mandarin into a shared clinical space, without parallel data or target-language fine-tuning. Evaluating 52 Mandarin speakers, contrastive alignment modestly outperforms the baseline (F1: 0.640 vs. 0.622) under leave-one-speaker-out evaluation. It also improves depressed-class recall at intermediate layers (7-8), though the small test set limits generalizability. Two findings remain robust: model scaling degrades cross-lingual performance while improving monolingual English, and speaker identity leakage artificially inflated previously reported Mandarin F1 scores to 0.954, an artifact we reproduce and quantify.

Prompt optimization improves language models without updating their weights by searching for a better system prompt, but its effectiveness varies widely across tasks. We study what makes a task amenable to prompt optimization. We show that the reward variance across different system prompts can be decomposed into two components: variance among responses, which captures generation stochasticity, and variance among system prompts, which captures differences in system prompt quality. Prompt optimization succeeds when variance among system prompts is sufficiently large, but fails when variance among responses dominates the variance of the system prompts. Surprisingly, we further show that scaling to more user prompts can hurt optimization by reducing variance among system prompts, especially on heterogeneous datasets where different user prompts favor different system prompts. Motivated by this insight, we propose p1, a simple user prompt filtering method that selects a small subset of user prompts with high variance across candidate system prompts. This subset of user prompts allows one to distinguish a good system prompt from a bad one, making system optimization easier. Experiments on reasoning benchmarks show that p1 substantially improves prompt optimization over training on the full dataset and outperforms strong baselines such as GEPA. Notably, training on only two prompts from AIME 24 yields a system prompt that generalizes well to other reasoning benchmarks.

Modern ML systems ingest data aggregated from diverse sources, such as synthetic, human-annotated, and live customer traffic. Understanding which examples are important to the performance of a learning algorithm is crucial for efficient model training. Recently, a growing body of literature has given rise to various "influence scores," which use training artifacts such as model confidence or checkpointed gradients to identify important subsets of data. However, these methods have primarily been developed in computer vision settings, and it remains unclear how well they generalize to language-based tasks using pretrained models. In this paper, we explore the applicability of influence scores in language classification tasks. We evaluate a diverse subset of these scores on the SNLI dataset by quantifying accuracy changes in response to pruning training data through random and influence-score-based sampling. We then stress-test one of the scores -- "variance of gradients" (VoG) from Agarwal et al. (2022) -- in an NLU model stack that was exposed to dynamic user speech patterns in a voice assistant type of setting. Our experiments demonstrate that in many cases, encoder-based language models can be finetuned on roughly 50% of the original data without degradation in performance metrics. Along the way, we summarize lessons learned from applying out-of-the-box implementations of influence scores, quantify the effects of noisy and class-imbalanced data, and offer recommendations on score-based sampling for better accuracy and training efficiency.

Ambient AI "scribe" systems promise to reduce clinical documentation burden, but automatic speech recognition (ASR) errors can remain unnoticed without careful review, and high-quality human reference transcripts are often unavailable for calibrating uncertainty. We investigate whether cross-model disagreement among heterogeneous ASR systems can act as a reference-free uncertainty signal to prioritize human verification in medical transcription workflows. Using 50 publicly available medical education audio clips (8 h 14 min), we transcribed each clip with eight ASR systems spanning commercial APIs and open-source engines. We aligned multi-model outputs, built consensus pseudo-references, and quantified token-level agreement using a majority-strength metric; we further characterized disagreements by type (content vs. punctuation/formatting) and assessed per-model agreement via leave-one-model-out (jackknife) consensus scoring. Inter-model reliability was low (ICC[2,1] = 0.131), indicating heterogeneous failure modes across systems. Across 76,398 evaluated token positions, 72.1% showed near-unanimous agreement (7-8 models), while 2.5% fell into high-risk bands (0-3 models), with high-risk mass varying from 0.7% to 11.4% across accent groups. Low-agreement regions were enriched for content disagreements, with the content fraction increasing from 53.9% to 73.9% across quintiles of high-risk mass. These results suggest that cross-model disagreement provides a sparse, localizable signal that can surface potentially unreliable transcript spans without human-verified references, enabling targeted review; clinical accuracy of flagged regions remains to be established.

Cross-speaker emotion intensity control aims to generate emotional speech of a target speaker with desired emotion intensities using only their neutral speech. A recently proposed method, emotion arithmetic, achieves emotion intensity control using a single-speaker emotion vector. Although this prior method has shown promising results in the same-speaker setting, it lost speaker consistency in the cross-speaker setting due to mismatches between the emotion vector of the source and target speakers. To overcome this limitation, we propose a speaker-agnostic emotion vector designed to capture shared emotional expressions across multiple speakers. This speaker-agnostic emotion vector is applicable to arbitrary speakers. Experimental results demonstrate that the proposed method succeeds in cross-speaker emotion intensity control while maintaining speaker consistency, speech quality, and controllability, even in the unseen speaker case.

Typical neural network trainings have substantial variance in test-set performance between repeated runs, impeding hyperparameter comparison and training reproducibility. We present the following results towards understanding this variation. (1) Despite having significant variance on their test-sets, we demonstrate that standard CIFAR-10 and ImageNet trainings have very little variance in their performance on the test-distributions from which those test-sets are sampled, suggesting that variance is less of a practical issue than previously thought. (2) We present a simplifying statistical assumption which closely approximates the structure of the test-set accuracy distribution. (3) We argue that test-set variance is inevitable in the following two senses. First, we show that variance is largely caused by high sensitivity of the training process to initial conditions, rather than by specific sources of randomness like the data order and augmentations. Second, we prove that variance is unavoidable given the observation that ensembles of trained networks are well-calibrated. (4) We conduct preliminary studies of distribution-shift, fine-tuning, data augmentation and learning rate through the lens of variance between runs.

Diffusion language models (DLMs) have emerged as a promising paradigm for large language models (LLMs), yet the non-deterministic behavior of DLMs remains poorly understood. The existing non-determinism evaluations for LLMs predominantly rely on dataset-level metrics under fixed inference configurations, providing limited insight into how model behavior varies across runs and evaluation conditions. In this work, we show that dataset-level metrics systematically attenuate non-determinism in diffusion language models by aggregating sample-level prediction quality across different runs. As a result, configurations with similar aggregate performance can exhibit substantially different behaviors on individual inputs, leaving fine-grained instability and distinct error patterns uncharacterized. To address this limitation, we conduct a fine-grained evaluation of non-determinism based on sample-level prediction differences across a range of model-related factors-including guidance scale, diffusion steps, and Monte Carlo sampling-as well as system-related factors such as batch size, hardware, and numerical precision. Our analysis reveals that non-determinism in DLMs is pervasive and structured, with code generation exhibiting markedly higher sensitivity to factor-level choices than question answering. To attribute sources of non-determinism evaluation, we introduce Factor Variance Attribution (FVA), a cross-factor analysis metric that decomposes observed non-determinism into variance attributable to different evaluation factor settings. Our findings highlight the need for fine-grained, factor-aware evaluation to enable reliable non-determinism assessment of diffusion language models.

An utterance-level speaker embedding is typically obtained by aggregating a sequence of frame-level representations. However, in real-world scenarios, individual frames encode not only speaker-relevant information but also various nuisance factors. As a result, different frames contribute unequally to the final utterance-level speaker representation for Automatic Speaker Verification systems. To address this issue, we propose to estimate the inherent uncertainty of each frame and assign adaptive weights accordingly, where frames with higher uncertainty receive lower attention. Based on this idea, we present U3-xi, a comprehensive framework designed to produce more reliable and interpretable uncertainty estimates for speaker embeddings. Specifically, we introduce several strategies for uncertainty supervision. First, we propose speaker-level uncertainty supervision via a Stochastic Variance Loss, where the distance between an utterance embedding and its corresponding speaker centroid serves as a pseudo ground truth for uncertainty learning. Second, we incorporate global-level uncertainty supervision by injecting the predicted uncertainty into the sof tmax scale during training. This adaptive scaling mechanism adjusts the sharpness of the decision boundary according to sample difficulty, providing global guidance. Third, we redesign the uncertainty estimation module by integrating a Transformer encoder with multi-view self-attention, enabling the model to capture rich local and long-range temporal dependencies. Comprehensive experiments demonstrate that U3-xi is model-agnostic and can be seamlessly applied to various speaker encoders. In particular, when applied to ECAPA-TDNN, it achieves 21.1% and 15.57% relative improvements on the VoxCeleb1 test sets in terms of EER and minDCF, respectively.

Modern Voice Cloning (VC) can synthesize speech that closely matches a target speaker from only seconds of reference audio, enabling applications such as personalized speech interfaces and dubbing. In practical deployments, modern audio generation models inevitably encounter noisy reference audios, imperfect text prompts, multilingual and long-form generation settings, downstream post-processing, and adversarial perturbations, all of which can significantly hurt robustness. Despite rapid progress in VC driven by autoregressive codec-token language models and diffusion-based models, robustness under realistic deployment shifts remains underexplored. This paper introduces RVCBench, a comprehensive dataset and benchmark that evaluates Robustness in Voice Clone. RVCBench contributes a large-scale, task-aligned robustness dataset that instantiates realistic deployment shifts through controlled text-audio pairing, multilingual and long-form scenarios, expressive prompts, post-processing conditions, and passive or proactive audio perturbations. Covering 18 robustness evaluations, 204 unique speakers, and 14,370 utterance-level evaluation items, RVCBench enables unified evaluation of input sensitivity, generation stability, output resilience, and perturbation robustness. We evaluate 18 representative modern open-source VC models and reveal systematic vulnerabilities in content consistency, speaker similarity, long-form stability, post-processing resilience, adversarial robustness, and detector-facing separability. We open-source the toolkit and dataset to support reproducible evaluation and future research.

Multi-media communications facilitate global interaction among people. However, despite researchers exploring cross-lingual translation techniques such as machine translation and audio speech translation to overcome language barriers, there is still a shortage of cross-lingual studies on visual speech. This lack of research is mainly due to the absence of datasets containing visual speech and translated text pairs. In this paper, we present AVMuST-TED, the first dataset for Audio-Visual Multilingual Speech Translation, derived from TED talks. Nonetheless, visual speech is not as distinguishable as audio speech, making it difficult to develop a mapping from source speech phonemes to the target language text. To address this issue, we propose MixSpeech, a cross-modality self-learning framework that utilizes audio speech to regularize the training of visual speech tasks. To further minimize the cross-modality gap and its impact on knowledge transfer, we suggest adopting mixed speech, which is created by interpolating audio and visual streams, along with a curriculum learning strategy to adjust the mixing ratio as needed. MixSpeech enhances speech translation in noisy environments, improving BLEU scores for four languages on AVMuST-TED by +1.4 to +4.2. Moreover, it achieves state-of-the-art performance in lip reading on CMLR (11.1\%), LRS2 (25.5\%), and LRS3 (28.0\%).

An ideal speech recognition model has the capability to transcribe speech accurately under various characteristics of speech signals, such as speaking style (read and spontaneous), speech context (formal and informal), and background noise conditions (clean and moderate). Building such a model requires a significant amount of training data with diverse speech characteristics. Currently, Indonesian data is dominated by read, formal, and clean speech, leading to a scarcity of Indonesian data with other speech variabilities. To develop Indonesian automatic speech recognition (ASR), we present our research on state-of-the-art speech recognition models, namely Massively Multilingual Speech (MMS) and Whisper, as well as compiling a dataset comprising Indonesian speech with variabilities to facilitate our study. We further investigate the models' predictive ability to transcribe Indonesian speech data across different variability groups. The best results were achieved by the Whisper fine-tuned model across datasets with various characteristics, as indicated by the decrease in word error rate (WER) and character error rate (CER). Moreover, we found that speaking style variability affected model performance the most.

Large language models (LLMs) are increasingly used as human simulators, both for evaluating conversational systems and for generating fine-tuning data. However, naive "act-as-a-user" prompting often yields verbose, unrealistic utterances, underscoring the need for principled evaluation of so-called user proxy agents. We present MIRRORBENCH, a reproducible, extensible benchmarking framework that evaluates user proxies solely on their ability to produce human-like user utterances across diverse conversational tasks, explicitly decoupled from downstream task success. MIRRORBENCH features a modular execution engine with typed interfaces, metadata-driven registries, multi-backend support, caching, and robust observability. The system supports pluggable user proxies, datasets, tasks, and metrics, enabling researchers to evaluate arbitrary simulators under a uniform, variance-aware harness. We include three lexical-diversity metrics (MATTR, YULE'S K, and HD-D) and three LLM-judge-based metrics (GTEval, Pairwise Indistinguishability, and Rubric-and-Reason). Across four open datasets, MIRRORBENCH yields variance-aware results and reveals systematic gaps between user proxies and real human users. The framework is open source and includes a simple command-line interface for running experiments, managing configurations and caching, and generating reports. The framework can be accessed at https://github.com/SAP/mirrorbench.

Retrieval-Augmented Generation (RAG) introduces a specific failure mode in discrete diffusion language models: when retrieved context contradicts parametric knowledge, the iterative denoising process becomes a visible battleground between competing knowledge sources. We identify temporal semantic divergence as an observable for detecting these conflicts and introduce the Trajectory Variance Score (TVS), a simple and interpretable measure of this divergence. TVS computes the mean pairwise cosine distance of answer embeddings across independent stochastic denoising trajectories, capturing the temporal tug of war between parametric and contextual attractors. Requiring as few as two parallel inference runs, TVS is computationally lightweight. Across four diverse datasets (Synthetic, SciQ, PopQA, and CounterFact), a simple Logistic Regression classifier using TVS achieves 70.10% accuracy and 0.7647 AUROC on LLaDA. On Dream 7B, increasing the number of trajectories from two to five improves accuracy from 63.91% to 69.62%. More complex sequential models provide only marginal improvements over the linear classifier. Evaluation across LLaDA and Dream 7B demonstrates that conflict-induced trajectory dynamics and their key properties transfer across distinct diffusion architectures.

Driven by advances in self-supervised learning for speech, state-of-the-art synthetic speech detectors have achieved low error rates on popular benchmarks such as ASVspoof. However, prior benchmarks do not address the wide range of real-world variability in speech. Are reported error rates realistic in real-world conditions? To assess detector failure modes and robustness under controlled distribution shifts, we introduce ShiftySpeech, a benchmark with more than 3000 hours of synthetic speech from 7 domains, 6 TTS systems, 12 vocoders, and 3 languages. We found that all distribution shifts degraded model performance, and contrary to prior findings, training on more vocoders, speakers, or with data augmentation did not guarantee better generalization. In fact, we found that training on less diverse data resulted in better generalization, and that a detector fit using samples from a single carefully selected vocoder and a single speaker achieved state-of-the-art results on the challenging In-the-Wild benchmark.

Automatic Speech Recognition (ASR) has become an integral component of modern technology, powering applications such as voice-activated assistants, transcription services, and accessibility tools. Yet ASR systems continue to struggle with the inherent variability of human speech, such as accents, dialects, and speaking styles, as well as environmental interference, including background noise. Moreover, domain-specific conversations often employ specialized terminology, which can exacerbate transcription errors. These shortcomings not only degrade raw ASR accuracy but also propagate mistakes through subsequent natural language processing pipelines. Because redesigning an ASR model is costly and time-consuming, non-intrusive refinement techniques that leave the model's architecture unchanged have become increasingly popular. In this survey, we systematically review current non-intrusive refinement approaches and group them into five classes: fusion, re-scoring, correction, distillation, and training adjustment. For each class, we outline the main methods, advantages, drawbacks, and ideal application scenarios. Beyond method classification, this work surveys adaptation techniques aimed at refining ASR in domain-specific contexts, reviews commonly used evaluation datasets along with their construction processes, and proposes a standardized set of metrics to facilitate fair comparisons. Finally, we identify open research gaps and suggest promising directions for future work. By providing this structured overview, we aim to equip researchers and practitioners with a clear foundation for developing more robust, accurate ASR refinement pipelines.

When audio and text conflict, speech-enabled language models follow text far more often than they do when arbitrating between two conflicting text sources, even under explicit instructions to trust the audio. We introduce ALME (Audio-LLM Modality Evaluation), a dataset of 57,602 controlled audio-text conflict stimuli across eight languages, together with Text Dominance Ratio (TDR), which measures how often a model follows conflicting text when instructed to follow audio. Gemini 2.0 Flash and GPT-4o show TDR 10--26times higher than a baseline that replaces audio with its transcript under otherwise identical conditions (Gemini 2.0 Flash: 16.6% vs. 1.6%; GPT-4o: 23.2% vs. 0.9%). These results suggest that text dominance reflects not only information content, but also an asymmetry in arbitration accessibility, i.e., how easily the model can use competing representations at decision time. Framing the transcript as deliberately corrupted reduces TDR by 80%, whereas forcing explicit transcription increases it by 14%. A fine-tuning ablation further suggests that arbitration behavior depends more on LLM reasoning than on the audio input path alone. Across four audio-LLMs, we observe the same qualitative pattern with substantial cross-model and cross-linguistic variation.

We introduce Vox-Profile, a comprehensive benchmark to characterize rich speaker and speech traits using speech foundation models. Unlike existing works that focus on a single dimension of speaker traits, Vox-Profile provides holistic and multi-dimensional profiles that reflect both static speaker traits (e.g., age, sex, accent) and dynamic speech properties (e.g., emotion, speech flow). This benchmark is grounded in speech science and linguistics, developed with domain experts to accurately index speaker and speech characteristics. We report benchmark experiments using over 15 publicly available speech datasets and several widely used speech foundation models that target various static and dynamic speaker and speech properties. In addition to benchmark experiments, we showcase several downstream applications supported by Vox-Profile. First, we show that Vox-Profile can augment existing speech recognition datasets to analyze ASR performance variability. Vox-Profile is also used as a tool to evaluate the performance of speech generation systems. Finally, we assess the quality of our automated profiles through comparison with human evaluation and show convergent validity. Vox-Profile is publicly available at: https://github.com/tiantiaf0627/vox-profile-release.

Agentic benchmarks increasingly rely on LLM-simulated users to scalably evaluate agent performance, yet the robustness, validity, and fairness of this approach remain unexamined. Through a user study with participants across the United States, India, Kenya, and Nigeria, we investigate whether LLM-simulated users serve as reliable proxies for real human users in evaluating agents on τ-Bench retail tasks. We find that user simulation lacks robustness, with agent success rates varying up to 9 percentage points across different user LLMs. Furthermore, evaluations using simulated users exhibit systematic miscalibration, underestimating agent performance on challenging tasks and overestimating it on moderately difficult ones. African American Vernacular English (AAVE) speakers experience consistently worse success rates and calibration errors than Standard American English (SAE) speakers, with disparities compounding significantly with age. We also find simulated users to be a differentially effective proxy for different populations, performing worst for AAVE and Indian English speakers. Additionally, simulated users introduce conversational artifacts and surface different failure patterns than human users. These findings demonstrate that current evaluation practices risk misrepresenting agent capabilities across diverse user populations and may obscure real-world deployment challenges.

Dialect differences caused by regional, social, and economic factors cause performance discrepancies for many groups of language technology users. Inclusive and equitable language technology must critically be dialect invariant, meaning that performance remains constant over dialectal shifts. Current systems often fall short of this ideal since they are designed and tested on a single dialect: Standard American English (SAE). We introduce a suite of resources for evaluating and achieving English dialect invariance. The resource is called Multi-VALUE, a controllable rule-based translation system spanning 50 English dialects and 189 unique linguistic features. Multi-VALUE maps SAE to synthetic forms of each dialect. First, we use this system to stress tests question answering, machine translation, and semantic parsing. Stress tests reveal significant performance disparities for leading models on non-standard dialects. Second, we use this system as a data augmentation technique to improve the dialect robustness of existing systems. Finally, we partner with native speakers of Chicano and Indian English to release new gold-standard variants of the popular CoQA task. To execute the transformation code, run model checkpoints, and download both synthetic and gold-standard dialectal benchmark datasets, see http://value-nlp.org.

Voice conversion aims to modify the source speaker's voice to resemble the target speaker while preserving the original speech content. Despite notable advancements in voice conversion these days, multi-lingual voice conversion (including both monolingual and cross-lingual scenarios) has yet to be extensively studied. It faces two main challenges: 1) the considerable variability in prosody and articulation habits across languages; and 2) the rarity of paired multi-lingual datasets from the same speaker. In this paper, we propose MulliVC, a novel voice conversion system that only converts timbre and keeps original content and source language prosody without multi-lingual paired data. Specifically, each training step of MulliVC contains three substeps: In step one the model is trained with monolingual speech data; then, steps two and three take inspiration from back translation, construct a cyclical process to disentangle the timbre and other information (content, prosody, and other language-related information) in the absence of multi-lingual data from the same speaker. Both objective and subjective results indicate that MulliVC significantly surpasses other methods in both monolingual and cross-lingual contexts, demonstrating the system's efficacy and the viability of the three-step approach with cycle consistency. Audio samples can be found on our demo page (mullivc.github.io).

Audio language models can recognize spoken commands and invoke tools, but an agent must first decide whether the acoustic and conversational context warrants action. We introduce VGBench, a 1,018-item diagnostic benchmark for action-level addressedness across side-talk, self-talk, and speaker-switch scenarios. Each item uses a shared action space comprising silence, a tool call, and a natural-language answer. Speaker-switch pairs hold the specified words fixed while source, distance rendering, and a temporal boundary define a controlled wearer-to-bystander shift. Six raw Audio LLMs and three training-free adaptations often identify the target tool yet rarely withhold action under this shift; the highest raw switch mute rate is 14%. We then use VoxGate as a post-training case study. Supervised training mutes 91.3% of switched commands while choosing the correct tool for all nearby wearer commands and text-only controls. An exploratory GRPO stage has similar switch performance; side-talk accuracy rises from 68.4% to 70.9%, and self-talk muting from 52.0% to 60.0%. Factorized controls identify an independent source-change effect, while sensitivity to the far-field manipulation varies across acoustic renderings. The benchmark therefore measures multi-cue acoustic-context gating rather than isolated speaker identity.

The development and evaluation of Large Language Models (LLMs) have largely focused on individual capabilities. However, this overlooks the intersection of multiple abilities across different types of expertise that are often required for real-world tasks, which we term cross capabilities. To systematically explore this concept, we first define seven core individual capabilities and then pair them to form seven common cross capabilities, each supported by a manually constructed taxonomy. Building on these definitions, we introduce CrossEval, a benchmark comprising 1,400 human-annotated prompts, with 100 prompts for each individual and cross capability. To ensure reliable evaluation, we involve expert annotators to assess 4,200 model responses, gathering 8,400 human ratings with detailed explanations to serve as reference examples. Our findings reveal that, in both static evaluations and attempts to enhance specific abilities, current LLMs consistently exhibit the "Law of the Weakest Link," where cross-capability performance is significantly constrained by the weakest component. Specifically, across 58 cross-capability scores from 17 models, 38 scores are lower than all individual capabilities, while 20 fall between strong and weak, but closer to the weaker ability. These results highlight the under-performance of LLMs in cross-capability tasks, making the identification and improvement of the weakest capabilities a critical priority for future research to optimize performance in complex, multi-dimensional scenarios.

We test and study the variation in speech recognition of fine-tuned versions of the Whisper model on child, elderly and non-native Dutch speech from the JASMIN-CGN corpus. Our primary goal is to evaluate how speakers' age and linguistic background influence Whisper's performance. Whisper achieves varying Word Error Rates (WER) when fine-tuned on subpopulations of specific ages and linguistic backgrounds. Fine-tuned performance is remarkably better than zero-shot performance, achieving a relative reduction in WER of 81% for native children, 72% for non-native children, 67% for non-native adults, and 65% for native elderly people. Our findings underscore the importance of training speech recognition models like Whisper on underrepresented subpopulations such as children, the elderly, and non-native speakers.

Despite rapid advancements in TTS models, a consistent and robust human evaluation framework is still lacking. For example, MOS tests fail to differentiate between similar models, and CMOS's pairwise comparisons are time-intensive. The MUSHRA test is a promising alternative for evaluating multiple TTS systems simultaneously, but in this work we show that its reliance on matching human reference speech unduly penalises the scores of modern TTS systems that can exceed human speech quality. More specifically, we conduct a comprehensive assessment of the MUSHRA test, focusing on its sensitivity to factors such as rater variability, listener fatigue, and reference bias. Based on our extensive evaluation involving 471 human listeners across Hindi and Tamil we identify two primary shortcomings: (i) reference-matching bias, where raters are unduly influenced by the human reference, and (ii) judgement ambiguity, arising from a lack of clear fine-grained guidelines. To address these issues, we propose two refined variants of the MUSHRA test. The first variant enables fairer ratings for synthesized samples that surpass human reference quality. The second variant reduces ambiguity, as indicated by the relatively lower variance across raters. By combining these approaches, we achieve both more reliable and more fine-grained assessments. We also release MANGO, a massive dataset of 47,100 human ratings, the first-of-its-kind collection for Indian languages, aiding in analyzing human preferences and developing automatic metrics for evaluating TTS systems.

We present NanoVoice, a personalized text-to-speech model that efficiently constructs voice adapters for multiple speakers simultaneously. NanoVoice introduces a batch-wise speaker adaptation technique capable of fine-tuning multiple references in parallel, significantly reducing training time. Beyond building separate adapters for each speaker, we also propose a parameter sharing technique that reduces the number of parameters used for speaker adaptation. By incorporating a novel trainable scale matrix, NanoVoice mitigates potential performance degradation during parameter sharing. NanoVoice achieves performance comparable to the baselines, while training 4 times faster and using 45 percent fewer parameters for speaker adaptation with 40 reference voices. Extensive ablation studies and analysis further validate the efficiency of our model.

Under strictly controlled pre-training settings, we observe a Crossover: when unique data is limited, diffusion language models (DLMs) consistently surpass autoregressive (AR) models by training for more epochs. The crossover shifts later with more or higher-quality data, earlier with larger models, and persists across dense and sparse architectures. We attribute the gains to three compounding factors: (1) any-order modeling, (2) super-dense compute from iterative bidirectional denoising, and (3) built-in Monte Carlo augmentation; input or parameter noise improves AR under data constraint but cannot close the gap. At scale, a 1.7B DLM trained with a ~1.5T-token compute budget on 10B unique Python tokens overtakes an AR coder trained with strictly matched settings. In addition, a 1B-parameter DLM achieves > 56% accuracy on HellaSwag and > 33% on MMLU using only 1B tokens, without any special tricks, just by repeating standard pre-training data. We also show that rising validation cross-entropy does not imply degraded downstream performance in this regime.

When workers lose jobs to AI-driven restructuring, two very different conversations happen on X (formerly Twitter) at the same time. Tech executives and AI researchers talk about productivity, transformation, and opportunity. Laid-off workers and labour critics talk about job loss, uncertainty, and fear. This paper asks a simple question: which conversation gets more reach? We report three studies using two collection methods and 763 tweets from 20 named public accounts. Study 1 used keyword-based collection (n=392) and found no significant difference between corpora (p=0.891), revealing that keyword search is too noisy for this task. Study 2 used account-based collection (n=96) and found a 3.12x mean amplification advantage for capital discourse over labour discourse (p=0.000003, Cohen's d=0.555). Study 3 combined both methods (n=763) and confirmed the finding at 4.18x mean and 10.77x median amplification ratio (p<0.000001). Critically, after normalising for follower count, the asymmetry persists at 2.69x (p=0.000009, Cohen's d=0.491), demonstrating that the effect is not simply a consequence of capital accounts having larger audiences. The finding is robust across all tested amplification metric weightings. We introduce the Amplification Ratio and Amplification Normalisation Index as simple metrics for measuring platform-level discourse inequality. A cross-platform replication on Reddit (n=647 posts) did not replicate the finding, suggesting the asymmetry may be specific to X's account-based amplification architecture. We discuss the methodological implications for cross-platform discourse analysis.

Automatic speech recognition (ASR) in non-English clinical settings is challenged by multiscript variability, where the same term may appear in multiple valid orthographic forms. Conventional string-matching evaluation metrics often underestimate ASR performance by treating orthographic variants as errors. To address this issue, we introduce MultiClin, a clinical ASR benchmark designed to evaluate robustness to multiscript variability. Experiments across diverse ASR models show that multiscript-aware evaluation provides a fairer assessment of recognition quality than conventional single-reference evaluation. We further investigate the impact of script consistency during training and find that inconsistent script mappings increase orthographic uncertainty and hinder model convergence, with a balanced 50% mapping ratio producing the highest entropy. In contrast, script unification consistently yields the best ASR performance. Our dataset and code are publicly available at: https://github.com/aitrics-ronaldo/Interspeech_MultiClin.

Multi-speaker text-to-speech (TTS) using a few adaption data is a challenge in practical applications. To address that, we propose a zero-shot multi-speaker TTS, named nnSpeech, that could synthesis a new speaker voice without fine-tuning and using only one adaption utterance. Compared with using a speaker representation module to extract the characteristics of new speakers, our method bases on a speaker-guided conditional variational autoencoder and can generate a variable Z, which contains both speaker characteristics and content information. The latent variable Z distribution is approximated by another variable conditioned on reference mel-spectrogram and phoneme. Experiments on the English corpus, Mandarin corpus, and cross-dataset proves that our model could generate natural and similar speech with only one adaption speech.

Cross-linguistic effects are a central topic in bilingual first-language acquisition. Artificial learners can help investigate L1-L2 interactions by enabling controlled comparisons across language combinations and learning conditions. Recent work explores this direction by training bilingual language models under developmentally plausible constraints. However, human and model learners still diverge in fundamental ways, with one major difference being input modality: children learn primarily from spoken input, whereas language models are typically trained on orthographic text. To reduce this gap, researchers have trained models on phonemic representations of speech. In this work, we combine these research directions to train bilingual BabyLMs with phonemic input. We keep English fixed as the L2 and vary the L1 across German, Swedish, Persian, and Basque, selected to represent contrasting combinations of syntactic and phoneme-inventory distance from English. Our results show stronger L1-related variation in grammatical learning trajectories under phonemic than orthographic input, while early lexical differences align with phoneme-inventory similarity.

Benchmarks measure whether a model is correct. They do not measure whether a model is reliable. This distinction is largely academic for single-shot inference, but becomes critical for agentic AI systems, where a single rephrased prompt can trigger cascading failures in multi-step execution. Yet this form of instability is not captured by existing evaluations. Hallucinations live in variance: they arise when semantically equivalent prompts activate inconsistent internal pathways, producing divergent outputs. Consistent but incorrect outputs reflect bias or missing knowledge; confident guessing reflects calibration failure. Neither constitutes hallucination under this definition. When error is variance-dominated, reducing redundant pathways improves reliability without adding knowledge. We formalize this through Semantic Stability (SS), measured via Paraphrase Consistency (PC@k): generate k paraphrases, greedy decode each, compute mode agreement. SS is a diagnostic for variance-driven unreliability, not a method for improving correctness. We show that a dense Qwen3-0.6B agrees with itself only 23.8% of the time; at 32% sparsity, agreement jumps to 55.9%. A phase diagram reveals the sweet spot where variance reduction outpaces bias accumulation, and regimes where stability collapses onto wrong answers.

While recent text-to-speech (TTS) systems increasingly integrate nonverbal vocalizations (NVs), their evaluations lack standardized metrics and reliable ground-truth references. To bridge this gap, we propose NV-Bench, the first benchmark grounded in a functional taxonomy that treats NVs as communicative acts rather than acoustic artifacts. NV-Bench comprises 1,651 multi-lingual, in-the-wild utterances with paired human reference audio, balanced across 14 NV categories. We introduce a dual-dimensional evaluation protocol: (1) Instruction Alignment, utilizing the proposed paralinguistic character error rate (PCER) to assess controllability, (2) Acoustic Fidelity, measuring the distributional gap to real recordings to assess acoustic realism. We evaluate diverse TTS models and develop two baselines. Experimental results demonstrate a strong correlation between our objective metrics and human perception, establishing NV-Bench as a standardized evaluation framework.

Human listeners readily adjust to unfamiliar speakers and language varieties through exposure, but do these adaptation benefits extend to state-of-the-art spoken language models? We introduce a scalable framework that allows for in-context learning (ICL) in Phi-4 Multimodal using interleaved task prompts and audio-text pairs, and find that as few as 12 example utterances (~50 seconds) at inference time reduce word error rates by a relative 19.7% (1.2 pp.) on average across diverse English corpora. These improvements are most pronounced in low-resource varieties, when the context and target speaker match, and when more examples are provided--though scaling our procedure yields diminishing marginal returns to context length. Overall, we find that our novel ICL adaptation scheme (1) reveals a similar performance profile to human listeners, and (2) demonstrates consistent improvements to automatic speech recognition (ASR) robustness across diverse speakers and language backgrounds. While adaptation succeeds broadly, significant gaps remain for certain varieties, revealing where current models still fall short of human flexibility. We release our prompts and code on GitHub.

While modern Text-to-Speech (TTS) systems can produce speech rated highly in terms of subjective evaluation, the distance between real and synthetic speech distributions remains understudied, where we use the term distribution to mean the sample space of all possible real speech recordings from a given set of speakers; or of the synthetic samples that could be generated for the same set of speakers. We evaluate the distance of real and synthetic speech distributions along the dimensions of the acoustic environment, speaker characteristics and prosody using a range of speech processing measures and the respective Wasserstein distances of their distributions. We reduce these distribution distances along said dimensions by providing utterance-level information derived from the measures to the model and show they can be generated at inference time. The improvements to the dimensions translate to overall distribution distance reduction approximated using Automatic Speech Recognition (ASR) by evaluating the fitness of the synthetic data as training data.

Cross-lingual evaluation of large language models (LLMs) typically conflates two sources of variance: genuine model performance differences and measurement instability. We investigate evaluation reliability by holding generation conditions constant while varying target language. Using synthetic customer-support dialogues generated with identical parameters across Estonian, Finnish, and Hungarian, we test whether automatic metrics and LLM-as-a-judge scoring produce stable model rankings across these morphologically rich, related Finno-Ugric languages. With a small set of Estonian native speaker annotations as a reference point, we find systematic ranking instabilities: surface-level metrics (lexical diversity, surface and semantic similarity) maintain cross-language stability, but pragmatic judgments (coherence, instruction-following) exhibit rank inversions and near-zero correlations. Because generation is controlled, these inconsistencies reflect how judge scoring behaves differently across languages rather than true model differences. This controlled design provides a diagnostic probe: evaluation methods that fail to maintain stability under identical generation conditions signal transfer failure before deployment. Our findings suggest that zero-shot judge transfer is unreliable for discourse-level assessment in morphologically rich languages, motivating language-specific calibration against targeted human baselines. We release our controlled generation protocol, synthetic data, and evaluation framework to enable replication across language families at https://github.com/isaac-chung/cross-lingual-stability-judges.

Audio-Visual Speech Recognition (AVSR) has achieved remarkable progress in offline conditions, yet its robustness in real-world video conferencing (VC) remains largely unexplored. This paper presents the first systematic evaluation of state-of-the-art AVSR models across mainstream VC platforms, revealing severe performance degradation caused by transmission distortions and spontaneous human hyper-expression. To address this gap, we construct MLD-VC, the first multimodal dataset tailored for VC, comprising 31 speakers, 22.79 hours of audio-visual data, and explicit use of the Lombard effect to enhance human hyper-expression. Through comprehensive analysis, we find that speech enhancement algorithms are the primary source of distribution shift, which alters the first and second formants of audio. Interestingly, we find that the distribution shift induced by the Lombard effect closely resembles that introduced by speech enhancement, which explains why models trained on Lombard data exhibit greater robustness in VC. Fine-tuning AVSR models on MLD-VC mitigates this issue, achieving an average 17.5% reduction in CER across several VC platforms. Our findings and dataset provide a foundation for developing more robust and generalizable AVSR systems in real-world video conferencing. MLD-VC is available at https://huggingface.co/datasets/nccm2p2/MLD-VC.

Synthesizing speech across different accents while preserving the speaker identity is essential for various real-world customer applications. However, the individual and accurate modeling of accents and speakers in a text-to-speech (TTS) system is challenging due to the complexity of accent variations and the intrinsic entanglement between the accent and speaker identity. In this paper, we present a novel approach for multi-speaker multi-accent TTS synthesis, which aims to synthesize voices of multiple speakers, each with various accents. Our proposed approach employs a multi-scale accent modeling strategy to address accent variations at different levels. Specifically, we introduce both global (utterance level) and local (phoneme level) accent modeling, supervised by individual accent classifiers to capture the overall variation within accented utterances and fine-grained variations between phonemes, respectively. To control accents and speakers separately, speaker-independent accent modeling is necessary, which is achieved by adversarial training with speaker classifiers to disentangle speaker identity within the multi-scale accent modeling. Consequently, we obtain speaker-independent and accent-discriminative multi-scale embeddings as comprehensive accent features. Additionally, we propose a local accent prediction model that allows to generate accented speech directly from phoneme inputs. Extensive experiments are conducted on an accented English speech corpus. Both objective and subjective evaluations show the superiority of our proposed system compared to baselines systems. Detailed component analysis demonstrates the effectiveness of global and local accent modeling, and speaker disentanglement on multi-speaker multi-accent speech synthesis.

Authorship attribution is the problem of identifying the most plausible author of an anonymous text from a set of candidate authors. Researchers have investigated same-topic and cross-topic scenarios of authorship attribution, which differ according to whether new, unseen topics are used in the testing phase. However, neither scenario allows us to explain whether errors are caused by a failure to capture authorship writing style or by a topic shift. Motivated by this, we propose the topic confusion task where we switch the author-topic configuration between the training and testing sets. This setup allows us to distinguish two types of errors: those caused by the topic shift and those caused by the features' inability to capture the writing styles. We show that stylometric features with part-of-speech tags are the least susceptible to topic variations. We further show that combining them with other features leads to significantly lower topic confusion and higher attribution accuracy. Finally, we show that pretrained language models such as BERT and RoBERTa perform poorly on this task and are surpassed by simple features such as word-level n-grams.

Data-parallel (DP) training with synchronous all-reduce is a dominant paradigm for full-parameter fine-tuning of large language models (LLMs). While parameter synchronization guarantees numerical equivalence of model weights after each iteration, it does not necessarily imply alignment of worker-level optimization dynamics before gradient aggregation. This paper identifies and studies this latent mismatch, termed silent inconsistency, where cross-worker divergence in losses and gradients can remain invisible under conventional aggregated monitoring signals. We propose a lightweight, model-agnostic diagnostic framework that quantifies worker-level consistency using training signals readily available in standard pipelines. Specifically, we introduce three complementary metrics: loss dispersion, gradient-norm dispersion, and gradient-direction consistency measured by inter-worker cosine similarity. The proposed metrics incur negligible overhead and require no modification to model architecture, synchronization mechanisms, or optimization algorithms. We validate the framework by fully fine-tuning the 1B-parameter openPangu-Embedded-1B-V1.1 model on the tatsu-lab/alpaca dataset using an 8-NPU DP setup, under controlled perturbations of cross-rank stochasticity. Experimental results show that progressively desynchronized data shuffling and random seeds lead to substantial increases in loss/gradient dispersion and reduced directional alignment, despite smooth globally averaged loss curves. These findings demonstrate that the proposed indicators provide actionable visibility into hidden instability modes in large-scale DP fine-tuning, enabling more reliable diagnosis and configuration assessment.

Speaker verification (SV) models are commonly assumed to better capture nuances among speaker characteristics as verification accuracy improves, leading to their widespread use as automated proxies for human voice similarity in speech generation tasks. However, by establishing a human perceptual alignment metric and conducting systematic analysis, we demonstrate that perceptual alignment is governed far more by how a model is trained (its learning objective) than by how well it performs (EER). Notably, standard margin-based classification losses (e.g., AAM-Softmax) yield substantially lower perceptual alignment than prototypical metric losses, while EER itself fails to track human judgment, directly challenging the community's implicit assumption. We trace this divergence to embedding geometry, where a model's effective dimensionality (d_{eff}) tracks perceptual alignment with a -0.95 rank correlation, revealing that the dimensional spread favored by classification losses fundamentally clashes with the low-dimensional nature of human voice perception. Imposing a dimensionality bottleneck compresses d_{eff} and raises perceptual alignment (ρ_{align}) from 0.08 to 0.74, establishing a principled geometric criterion for evaluating voice similarity.

In this paper, we introduce a zero-shot Voice Transfer (VT) module that can be seamlessly integrated into a multi-lingual Text-to-speech (TTS) system to transfer an individual's voice across languages. Our proposed VT module comprises a speaker-encoder that processes reference speech, a bottleneck layer, and residual adapters, connected to preexisting TTS layers. We compare the performance of various configurations of these components and report Mean Opinion Score (MOS) and Speaker Similarity across languages. Using a single English reference speech per speaker, we achieve an average voice transfer similarity score of 73% across nine target languages. Vocal characteristics contribute significantly to the construction and perception of individual identity. The loss of one's voice, due to physical or neurological conditions, can lead to a profound sense of loss, impacting one's core identity. As a case study, we demonstrate that our approach can not only transfer typical speech but also restore the voices of individuals with dysarthria, even when only atypical speech samples are available - a valuable utility for those who have never had typical speech or banked their voice. Cross-lingual typical audio samples, plus videos demonstrating voice restoration for dysarthric speakers are available here (google.github.io/tacotron/publications/zero_shot_voice_transfer).

Parameter-efficient fine-tuning (PEFT) for personalizing automatic speech recognition (ASR) has recently shown promise for adapting general population models to atypical speech. However, these approaches assume a priori knowledge of the atypical speech disorder being adapted for -- the diagnosis of which requires expert knowledge that is not always available. Even given this knowledge, data scarcity and high inter/intra-speaker variability further limit the effectiveness of traditional fine-tuning. To circumvent these challenges, we first identify the minimal set of model parameters required for ASR adaptation. Our analysis of each individual parameter's effect on adaptation performance allows us to reduce Word Error Rate (WER) by half while adapting 0.03% of all weights. Alleviating the need for cohort-specific models, we next propose the novel use of a meta-learned hypernetwork to generate highly individualized, utterance-level adaptations on-the-fly for a diverse set of atypical speech characteristics. Evaluating adaptation at the global, cohort and individual-level, we show that hypernetworks generalize better to out-of-distribution speakers, while maintaining an overall relative WER reduction of 75.2% using 0.1% of the full parameter budget.

Full-duplex evaluation often emphasizes whether an agent keeps speaking or stops. That binary cannot express a third response humans use routinely: continuing to speak while incorporating what the listener just contributed. The contribution may be a missing word, a correction or a clarification. We introduce Duplex Cue, an evaluation of this in-turn adaptation in full-duplex voice agents. Duplex Cue separates listener intent (backchannel, collaboration, or interruption) from speaker behavior: continuing unchanged, adapting within the turn, or yielding. Adaptation includes acknowledgment as well as content revision. In a single-model case study using 300 human-confirmed cues from unscripted English conversations, we compare recorded human responses with PersonaPlex continuations generated while replaying the listener's audio. We retain 208 pairs with the ongoing speaker active at cue onset and a scorable response in each condition. On the 66 collaborative pairs, recorded speakers adapt in 68.2\% of cases, compared with 34.8\% for PersonaPlex. The model otherwise continues unchanged (42.4\%) or yields (22.7\%). These findings show why evaluating natural voice interaction requires measuring how an agent responds to a listener's contribution as well as whether it keeps speaking.

Recent breakthroughs in multi-talker ASR (MT-ASR) and speaker diarization (SD) rely on synthetic data to mitigate the scarcity of large-scale conversational recordings, yet the impact of specific simulation choices remains poorly understood. To mind the gap between simulated mixtures and real-world interactions, we present a study of synthetic data generation for leading MT-ASR (DiCoW) and SD (Sortformer) systems. By introducing FastMSS, a highly efficient open-source simulator, we analyze turn-taking dynamics, source domain, acoustic augmentation, and data mixing strategies. Our findings reveal that optimal simulation recipes are highly task-dependent: increasing speech overlap benefits ASR but degrades diarization. Furthermore, broad source diversity consistently outperforms exact domain matching. Ultimately, synthetic-only training approaches real-data baselines, and combining simulated data with real recordings yields substantial gains over real-only training across both tasks.

Most automatic speaker verification (ASV) systems operate on individual utterances, despite real-world interactions typically consisting of multiple utterances. As speech accumulates, increasingly rich speaker information becomes available through acoustic, prosodic, and linguistic cues, potentially challenging speaker anonymization methods that primarily target vocal characteristics. We investigate ASV in a multi-utterance, multimodal setting and examine whether aggregating information across anonymized speech impacts privacy. We first study audio-only aggregation across multiple anonymized utterances and observe consistent performance improvements as more speech becomes available. We then incorporate prosodic and linguistic information, showing that multimodal systems outperform unimodal approaches. Finally, we compare aggregation strategies and find that frame-level aggregation yields the lowest EERs. Even with only five anonymized utterances, combining audio and text reduces EER by over 15% relative to audio-only aggregation, demonstrating that substantial speaker-discriminative information remains accessible despite anonymization.

Autoregressive neural codec language models have shown strong zero-shot voice cloning ability, but decoder-only architectures treat input text as a prefix that competes with the growing audio sequence for positional capacity, weakening text conditioning over long utterances. We present T5Gemma-TTS, an encoder-decoder codec language model that maintains persistent text conditioning by routing bidirectional text representations through cross-attention at every decoder layer. Built on the T5Gemma pretrained encoder-decoder backbone (2B encoder + 2B decoder; 4B parameters), it inherits rich linguistic knowledge without phoneme conversion and processes text directly at the subword level. To improve duration control, we introduce Progress-Monitoring Rotary Position Embedding (PM-RoPE) in all 26 cross-attention layers, injecting normalized progress signals that help the decoder track target speech length. Trained on 170,000 hours of multilingual speech in English, Chinese, and Japanese, T5Gemma-TTS achieves a statistically significant speaker-similarity gain on Japanese over XTTSv2 (0.677 vs. 0.622; non-overlapping 95% confidence intervals) and the highest numerical Korean speaker similarity (0.747) despite Korean not being included in training, although this margin over XTTSv2 (0.741) is not statistically conclusive. It also attains the lowest numerical Japanese character error rate among five baselines (0.126), though this ranking should be interpreted cautiously because of partial confidence-interval overlap with Kokoro. English results on LibriSpeech should be viewed as an upper-bound estimate because LibriHeavy is a superset of LibriSpeech. Using the same checkpoint, disabling PM-RoPE at inference causes near-complete synthesis failure: CER degrades from 0.129 to 0.982 and duration accuracy drops from 79% to 46%. Code and weights are available at https://github.com/Aratako/T5Gemma-TTS.

Code-switching (CS) is the alternating use of two or more languages within a conversation or utterance, often influenced by social context and speaker identity. This linguistic phenomenon poses challenges for Automatic Speech Recognition (ASR) systems, which are typically designed for a single language and struggle to handle multilingual inputs. The growing global demand for multilingual applications, including Code-Switching ASR (CSASR), Text-to-Speech (CSTTS), and Cross-Lingual Information Retrieval (CLIR), highlights the inadequacy of existing monolingual datasets. Although some code-switching datasets exist, most are limited to bilingual mixing within homogeneous ethnic groups, leaving a critical need for a large-scale, diverse benchmark akin to ImageNet in computer vision. To bridge this gap, we introduce LinguaMaster, a multi-agent collaboration framework specifically designed for efficient and scalable multilingual data synthesis. Leveraging this framework, we curate SwitchLingua, the first large-scale multilingual and multi-ethnic code-switching dataset, including: (1) 420K CS textual samples across 12 languages, and (2) over 80 hours of audio recordings from 174 speakers representing 18 countries/regions and 63 racial/ethnic backgrounds, based on the textual data. This dataset captures rich linguistic and cultural diversity, offering a foundational resource for advancing multilingual and multicultural research. Furthermore, to address the issue that existing ASR evaluation metrics lack sensitivity to code-switching scenarios, we propose the Semantic-Aware Error Rate (SAER), a novel evaluation metric that incorporates semantic information, providing a more accurate and context-aware assessment of system performance.

Recent innovations on hardware (e.g. Nvidia A100) have motivated learning N:M structured sparsity masks from scratch for fast model inference. However, state-of-the-art learning recipes in this regime (e.g. SR-STE) are proposed for non-adaptive optimizers like momentum SGD, while incurring non-trivial accuracy drop for Adam-trained models like attention-based LLMs. In this paper, we first demonstrate such gap origins from poorly estimated second moment (i.e. variance) in Adam states given by the masked weights. We conjecture that learning N:M masks with Adam should take the critical regime of variance estimation into account. In light of this, we propose STEP, an Adam-aware recipe that learns N:M masks with two phases: first, STEP calculates a reliable variance estimate (precondition phase) and subsequently, the variance remains fixed and is used as a precondition to learn N:M masks (mask-learning phase). STEP automatically identifies the switching point of two phases by dynamically sampling variance changes over the training trajectory and testing the sample concentration. Empirically, we evaluate STEP and other baselines such as ASP and SR-STE on multiple tasks including CIFAR classification, machine translation and LLM fine-tuning (BERT-Base, GPT-2). We show STEP mitigates the accuracy drop of baseline recipes and is robust to aggressive structured sparsity ratios.

Speech conveys more information than just text, as the same word can be uttered in various voices to convey diverse information. Compared to traditional text-to-speech (TTS) methods relying on speech prompts (reference speech) for voice variability, using text prompts (descriptions) is more user-friendly since speech prompts can be hard to find or may not exist at all. TTS approaches based on the text prompt face two challenges: 1) the one-to-many problem, where not all details about voice variability can be described in the text prompt, and 2) the limited availability of text prompt datasets, where vendors and large cost of data labeling are required to write text prompt for speech. In this work, we introduce PromptTTS 2 to address these challenges with a variation network to provide variability information of voice not captured by text prompts, and a prompt generation pipeline to utilize the large language models (LLM) to compose high quality text prompts. Specifically, the variation network predicts the representation extracted from the reference speech (which contains full information about voice) based on the text prompt representation. For the prompt generation pipeline, it generates text prompts for speech with a speech understanding model to recognize voice attributes (e.g., gender, speed) from speech and a large language model to formulate text prompt based on the recognition results. Experiments on a large-scale (44K hours) speech dataset demonstrate that compared to the previous works, PromptTTS 2 generates voices more consistent with text prompts and supports the sampling of diverse voice variability, thereby offering users more choices on voice generation. Additionally, the prompt generation pipeline produces high-quality prompts, eliminating the large labeling cost. The demo page of PromptTTS 2 is available onlinehttps://speechresearch.github.io/prompttts2.

Voice provides a natural and immediate interface for AI agents. Many settings in which voice agents could be useful, including meetings, households, and collaborative work, are inherently multi-speaker. Supporting these settings introduces challenges that are largely absent from one-on-one interaction. We introduce the Multi-Speaker Interaction Benchmark (MSI-Bench) for evaluating multi-speaker voice interaction. Each test case is a short multi-party multi-turn audio scene with participant context, expected tool calls, and atomic rubrics. The benchmark targets three capability families: multi-speaker memory, multi-speaker instruction following, and multi-speaker reasoning. It comprises 1,152 test cases, evenly split between Mandarin Chinese and English (576 each). The strongest configuration on each split passes all rubrics on only 66.8% of English and 54.5% of Mandarin cases, and the strongest open-weight configuration on 34.0% and 19.3%. Failure analysis separates perception from reasoning: open-weight models are bottlenecked by the multi-speaker audio front-end, while frontier systems still fail speaker-scoped decision making on clean transcripts---and models across the board often respond when no one has addressed them. These results identify speaker-grounded perception, speaker-scoped decision making, and conversational restraint as concrete targets for future voice agents.

India's 22 scheduled languages are spoken by over a billion people in conversations that mix languages, often within one turn, overlap freely, and take place over telephones, in meetings, and in media. Knowing who spoke when, the task of speaker diarization, is essential to speech applications for such conversation: voice agents conversing with several people at once, speaker attributed transcription of meetings and calls, contact center analytics, and media monitoring. We present a systematic study of speaker diarization across all 22 languages on IndicDiarBench and introduce Anuvani-Diarization, which adapts a self-supervised end-to-end segmentation model to Indian conversation with uncertainty aware silver supervision. Evaluated zero shot alongside nine recent diarization systems, it reaches a duration weighted diarization error rate (DER) of 16.78% on all 1,164 recordings, the lowest of all systems we ran and second only to the best published result (16.0%), and leads the three strongest rivals in 14 of 22 languages. It matches its parent on six international benchmarks. We also show that the benchmark's utterance level annotation convention is a hidden variable worth 1.9 to 4.1 DER points, depending on the system, and that a convention learned from other Indian transcriptions does not transfer.

We propose to address online speaker diarization as a combination of incremental clustering and local diarization applied to a rolling buffer updated every 500ms. Every single step of the proposed pipeline is designed to take full advantage of the strong ability of a recently proposed end-to-end overlap-aware segmentation to detect and separate overlapping speakers. In particular, we propose a modified version of the statistics pooling layer (initially introduced in the x-vector architecture) to give less weight to frames where the segmentation model predicts simultaneous speakers. Furthermore, we derive cannot-link constraints from the initial segmentation step to prevent two local speakers from being wrongfully merged during the incremental clustering step. Finally, we show how the latency of the proposed approach can be adjusted between 500ms and 5s to match the requirements of a particular use case, and we provide a systematic analysis of the influence of latency on the overall performance (on AMI, DIHARD and VoxConverse).

Accent variability remains a major errors in automatic speech recognition, yet most adaptation methods rely on parameter fine-tuning without understanding where accent information is encoded. We treat accent variation as an interpretable subspace in hidden representations and investigate whether it can be identified and controlled directly in activation space. We extract layer-wise encoder activations and estimate mean-shift directions capturing accent-induced representation shifts. By injecting these directions into individual layers and measuring how they align accented and standard embeddings, we derive a layer-wise accent sensitivity profile, revealing that accent information concentrates in a narrow band of middle encoder layers. Leveraging this structure, we further introduce parameter-free accent steering that modifies representations during inference without updating model weights. Experiments across eight accents show consistent word error rate reductions.

Natural language processing (NLP) and speech technologies have made significant progress in recent years; however, they remain largely focused on standardized language varieties. Dialects, despite their cultural significance and widespread use, are underrepresented in linguistic resources and computational models, resulting in performance disparities. To address this gap, we introduce Saar-Voice, a six-hour speech corpus for the Saarbrücken dialect of German. The dataset was created by first collecting text through digitized books and locally sourced materials. A subset of this text was recorded by nine speakers, and we conducted analyses on both the textual and speech components to assess the dataset's characteristics and quality. We discuss methodological challenges related to orthographic and speaker variation, and explore grapheme-to-phoneme (G2P) conversion. The resulting corpus provides aligned textual and audio representations. This serves as a foundation for future research on dialect-aware text-to-speech (TTS), particularly in low-resource scenarios, including zero-shot and few-shot model adaptation.

Voice agents, artificial intelligence systems that conduct spoken conversations to complete tasks, are increasingly deployed across enterprise applications. However, no existing benchmark jointly addresses two core evaluation challenges: generating realistic simulated conversations, and measuring quality across the full scope of voice-specific failure modes. We present EVA-Bench, an end-to-end evaluation framework that addresses both. On the simulation side, EVA-Bench orchestrates bot-to-bot audio conversations over dynamic multi-turn dialogues, with automatic simulation validation that detects user simulator error and appropriately regenerates conversations before scoring. On the measurement side, EVA-Bench introduces two composite metrics: EVA-A (Accuracy), capturing task completion, faithfulness, and audio-level speech fidelity; and EVA-X (Experience), capturing conversation progression, spoken conciseness, and turn-taking timing. Both metrics apply to different agent architectures, enabling direct cross-architecture comparison. EVA-Bench includes 213 scenarios across three enterprise domains, a controlled perturbation suite for accent and noise robustness, and pass@1, pass@k, pass^k measurements that distinguish peak from reliable capability. Across 12 systems spanning all three architectures, we find: (1) no system simultaneously exceeds 0.5 on both EVA-A pass@1 and EVA-X pass@1; (2) peak and reliable performance diverge substantially (median pass@k - pass^k gap of 0.44 on EVA-A); and (3) accent and noise perturbations expose substantial robustness gaps, with effects varying across architectures, systems, and metrics (mean up to 0.314). We release the full framework, evaluation suite, and benchmark data under an open-source license.

Large language models often remain sensitive to answer format: a question solved correctly in one form may fail in another semantically equivalent form. To study this gap, we define cross-format robustness as the extent to which a model answers the same underlying question consistently across formats. We then compare full-format training with FormatMix, which expands only a subset of training items into multiple equivalent formats using either random or targeted selection. Across GLM4 and Llama-3.1, multi-format supervision consistently improves both task performance and cross-format robustness, whereas Multiple-choice question (MCQ)-only supervision alone brings little benefit and can even reduce robustness. We further find that expanding only about 30% of the training set into multiple formats often recovers most of the gain from full-format training, and this effect appears across the model families and sizes we study. These results suggest that format diversity, rather than additional supervision alone, is the key driver of robustness. That lightweight multi-format augmentation is a practical way to make LLMs less sensitive to answer format without changing the base model.

Large Language Models (LLMs) have demonstrated impressive generative capabilities across diverse tasks but remain susceptible to hallucinations, confidently generated yet factually incorrect outputs. We introduce a reference-free, token-level hallucination detection framework that leverages the variance in token log-probabilities across multiple stochastic generations. Unlike prior methods that require ground-truth references or sentence-level verification, our approach is model-agnostic, interpretable, and suited for real-time or post-hoc analysis. We evaluate our method on unanswerable question prompts from the SQuAD v2 dataset and benchmark across three autoregressive models of varying scales: GPT-Neo 125M, Falcon 1B, and Mistral 7B. Through both quantitative metrics and visual diagnostics, we show that token-level variance reliably highlights instability in model outputs and correlates with hallucination patterns. Our framework is lightweight, reproducible, and adaptable to multiple domains, offering a valuable diagnostic tool for analyzing generative reliability in LLMs.

Research on cross-dialectal transfer from a standard to a non-standard dialect variety has typically focused on text data. However, dialects are primarily spoken, and non-standard spellings are known to cause issues in text processing. We compare standard-to-dialect transfer in three settings: text models, speech models, and cascaded systems where speech first gets automatically transcribed and then further processed by a text model. In our experiments, we focus on German and multiple German dialects in the context of written and spoken intent and topic classification. To that end, we release the first dialectal audio intent classification dataset. We find that the speech-only setup provides the best results on the dialect data while the text-only setup works best on the standard data. While the cascaded systems lag behind the text-only models for German, they perform relatively well on the dialectal data if the transcription system generates normalized, standard-like output.

This paper describes the Microsoft end-to-end neural text to speech (TTS) system: DelightfulTTS for Blizzard Challenge 2021. The goal of this challenge is to synthesize natural and high-quality speech from text, and we approach this goal in two perspectives: The first is to directly model and generate waveform in 48 kHz sampling rate, which brings higher perception quality than previous systems with 16 kHz or 24 kHz sampling rate; The second is to model the variation information in speech through a systematic design, which improves the prosody and naturalness. Specifically, for 48 kHz modeling, we predict 16 kHz mel-spectrogram in acoustic model, and propose a vocoder called HiFiNet to directly generate 48 kHz waveform from predicted 16 kHz mel-spectrogram, which can better trade off training efficiency, modelling stability and voice quality. We model variation information systematically from both explicit (speaker ID, language ID, pitch and duration) and implicit (utterance-level and phoneme-level prosody) perspectives: 1) For speaker and language ID, we use lookup embedding in training and inference; 2) For pitch and duration, we extract the values from paired text-speech data in training and use two predictors to predict the values in inference; 3) For utterance-level and phoneme-level prosody, we use two reference encoders to extract the values in training, and use two separate predictors to predict the values in inference. Additionally, we introduce an improved Conformer block to better model the local and global dependency in acoustic model. For task SH1, DelightfulTTS achieves 4.17 mean score in MOS test and 4.35 in SMOS test, which indicates the effectiveness of our proposed system

Recent work has explored sequence-to-sequence latent variable models for expressive speech synthesis (supporting control and transfer of prosody and style), but has not presented a coherent framework for understanding the trade-offs between the competing methods. In this paper, we propose embedding capacity (the amount of information the embedding contains about the data) as a unified method of analyzing the behavior of latent variable models of speech, comparing existing heuristic (non-variational) methods to variational methods that are able to explicitly constrain capacity using an upper bound on representational mutual information. In our proposed model (Capacitron), we show that by adding conditional dependencies to the variational posterior such that it matches the form of the true posterior, the same model can be used for high-precision prosody transfer, text-agnostic style transfer, and generation of natural-sounding prior samples. For multi-speaker models, Capacitron is able to preserve target speaker identity during inter-speaker prosody transfer and when drawing samples from the latent prior. Lastly, we introduce a method for decomposing embedding capacity hierarchically across two sets of latents, allowing a portion of the latent variability to be specified and the remaining variability sampled from a learned prior. Audio examples are available on the web.

Globalization and multiculturalism continue to produce increasingly diverse speech varieties. Yet current spoken dialogue systems frequently fail on under-represented dialects and accents, often misidentifying the input language and causing cascading failures in downstream dialogue tasks. Addressing this dialectal variance under low-resource constraints remains an open challenge, as standard fine-tuning is computationally expensive and prone to overfitting on high-dimensional speech data. We propose Convex Language Detection (CLD), a novel framework that integrates theoretically grounded convex optimization techniques into the spoken dialogue systems pipeline. Our method is efficiently implemented via multi-GPU Alternating Direction Method of Multipliers (ADMM) in JAX, thus providing global optimality guarantees and fast training in polynomial time. Theoretically, we prove that our convex objective induces certified margin stability and provide guarantees against feature perturbations. Empirically, we demonstrate sample efficiency and robustness to input dialectical variation, achieving 97-98% accuracy in challenging low-resource regimes. Our open-source package is available at https://pypi.org/project/jaxcld/

Spoken query retrieval is an important interaction mode in modern information retrieval. However, existing evaluation datasets are often limited to simple queries under constrained noise conditions, making them inadequate for assessing the robustness of spoken query retrieval systems under complex acoustic perturbations. To address this limitation, we present SQuTR, a robustness benchmark for spoken query retrieval that includes a large-scale dataset and a unified evaluation protocol. SQuTR aggregates 37,317 unique queries from six commonly used English and Chinese text retrieval datasets, spanning multiple domains and diverse query types. We synthesize speech using voice profiles from 200 real speakers and mix 17 categories of real-world environmental noise under controlled SNR levels, enabling reproducible robustness evaluation from quiet to highly noisy conditions. Under the unified protocol, we conduct large-scale evaluations on representative cascaded and end-to-end retrieval systems. Experimental results show that retrieval performance decreases as noise increases, with substantially different drops across systems. Even large-scale retrieval models struggle under extreme noise, indicating that robustness remains a critical bottleneck. Overall, SQuTR provides a reproducible testbed for benchmarking and diagnostic analysis, and facilitates future research on robustness in spoken query to text retrieval.

We describe a neural network-based system for text-to-speech (TTS) synthesis that is able to generate speech audio in the voice of many different speakers, including those unseen during training. Our system consists of three independently trained components: (1) a speaker encoder network, trained on a speaker verification task using an independent dataset of noisy speech from thousands of speakers without transcripts, to generate a fixed-dimensional embedding vector from seconds of reference speech from a target speaker; (2) a sequence-to-sequence synthesis network based on Tacotron 2, which generates a mel spectrogram from text, conditioned on the speaker embedding; (3) an auto-regressive WaveNet-based vocoder that converts the mel spectrogram into a sequence of time domain waveform samples. We demonstrate that the proposed model is able to transfer the knowledge of speaker variability learned by the discriminatively-trained speaker encoder to the new task, and is able to synthesize natural speech from speakers that were not seen during training. We quantify the importance of training the speaker encoder on a large and diverse speaker set in order to obtain the best generalization performance. Finally, we show that randomly sampled speaker embeddings can be used to synthesize speech in the voice of novel speakers dissimilar from those used in training, indicating that the model has learned a high quality speaker representation.

Prosody diversity is essential for achieving naturalness and expressiveness in zero-shot text-to-speech (TTS). However, frequently used acoustic metrics capture only partial views of prosodic variation and correlate poorly with human perception, leaving the problem of reliably quantifying prosody diversity underexplored. To bridge this gap, we introduce ProsodyEval, a prosody diversity assessment dataset that provides Prosody Mean Opinion Score (PMOS) alongside conventional acoustic metrics. ProsodyEval comprises 1000 speech samples derived from 7 mainstream TTS systems, with 2000 human ratings. Building on this, we propose the Discretized Speech Weighted Edit Distance (DS-WED), a new objective diversity metric that quantifies prosodic variation via weighted edit distance over semantic tokens. Experiments on ProsodyEval show that DS-WED achieves substantially higher correlation with human judgments than existing acoustic metrics, while remaining highly robust in speech tokenization from HuBERT and WavLM. Leveraging DS-WED, we benchmark state-of-the-art open-source TTS systems on LibriSpeech test-clean and Seed-TTS test-en, and further explorations uncover several factors that influence prosody diversity, including generative modeling paradigms, duration control, and reinforcement learning. Moreover, we find that current large audio language models (LALMs) remain limited in capturing prosodic variations. Audio samples are available at https://prosodyeval.github.io.

Research on Large Language Models (LLMs) studies output variation across generation, reasoning, alignment, and representational analysis, often under the umbrella of "diversity." Yet the terminology remains fragmented, largely because the normative objectives underlying tasks are rarely made explicit. We introduce the Magic, Madness, Heaven, Sin framework, which models output variation along a homogeneity-heterogeneity axis, where valuation is determined by the task and its normative objective. We organize tasks into four normative contexts: epistemic (factuality), interactional (user utility), societal (representation), and safety (robustness). For each, we examine the failure modes and vocabulary such as hallucination, mode collapse, bias, and erasure through which variation is studied. We apply the framework to analyze all pairwise cross-contextual interactions, revealing that optimizing for one objective, such as improving safety, can inadvertently harm demographic representation or creative diversity. We argue for context-aware evaluation of output variation, reframing it as a property shaped by task objectives rather than a model's intrinsic trait.

Existing video personalization methods preserve visual likeness but treat video and audio separately. Without access to the visual scene, audio models cannot synchronize sounds with on-screen actions; and because classical voice-cloning models condition only on a reference recording, a text prompt cannot redirect speaking style or acoustic environment. We propose ID-LoRA (Identity-Driven In-Context LoRA), which jointly generates a subject's appearance and voice in a single model, letting a text prompt, a reference image, and a short audio clip govern both modalities together. ID-LoRA adapts the LTX-2 joint audio-video diffusion backbone via parameter-efficient In-Context LoRA and, to our knowledge, is the first method to personalize visual appearance and voice in a single generative pass. Two challenges arise. Reference and generation tokens share the same positional-encoding space, making them hard to distinguish; we address this with negative temporal positions, placing reference tokens in a disjoint RoPE region while preserving their internal temporal structure. Speaker characteristics also tend to be diluted during denoising; we introduce identity guidance, a classifier-free guidance variant that amplifies speaker-specific features by contrasting predictions with and without the reference signal. In human preference studies, ID-LoRA is preferred over Kling 2.6 Pro by 73% of annotators for voice similarity and 65% for speaking style. On cross-environment settings, speaker similarity improves by 24% over Kling, with the gap widening as conditions diverge. A preliminary user study further suggests that joint generation provides a useful inductive bias for physically grounded sound synthesis. ID-LoRA achieves these results with only ~3K training pairs on a single GPU. Code, models, and data will be released.

Speech-to-speech translation systems today do not adequately support use for dialog purposes. In particular, nuances of speaker intent and stance can be lost due to improper prosody transfer. We present an exploration of what needs to be done to overcome this. First, we developed a data collection protocol in which bilingual speakers re-enact utterances from an earlier conversation in their other language, and used this to collect an English-Spanish corpus, so far comprising 1871 matched utterance pairs. Second, we developed a simple prosodic dissimilarity metric based on Euclidean distance over a broad set of prosodic features. We then used these to investigate cross-language prosodic differences, measure the likely utility of three simple baseline models, and identify phenomena which will require more powerful modeling. Our findings should inform future research on cross-language prosody and the design of speech-to-speech translation systems capable of effective prosody transfer.

Current state of the art acoustic models can easily comprise more than 100 million parameters. This growing complexity demands larger training datasets to maintain a decent generalization of the final decision function. An ideal dataset is not necessarily large in size, but large with respect to the amount of unique speakers, utilized hardware and varying recording conditions. This enables a machine learning model to explore as much of the domain-specific input space as possible during parameter estimation. This work introduces Common Phone, a gender-balanced, multilingual corpus recorded from more than 11.000 contributors via Mozilla's Common Voice project. It comprises around 116 hours of speech enriched with automatically generated phonetic segmentation. A Wav2Vec 2.0 acoustic model was trained with the Common Phone to perform phonetic symbol recognition and validate the quality of the generated phonetic annotation. The architecture achieved a PER of 18.1 % on the entire test set, computed with all 101 unique phonetic symbols, showing slight differences between the individual languages. We conclude that Common Phone provides sufficient variability and reliable phonetic annotation to help bridging the gap between research and application of acoustic models.

In this paper, we describe SpeakerStew - a hybrid system to perform speaker verification on 46 languages. Two core ideas were explored in this system: (1) Pooling training data of different languages together for multilingual generalization and reducing development cycles; (2) A novel triage mechanism between text-dependent and text-independent models to reduce runtime cost and expected latency. To the best of our knowledge, this is the first study of speaker verification systems at the scale of 46 languages. The problem is framed from the perspective of using a smart speaker device with interactions consisting of a wake-up keyword (text-dependent) followed by a speech query (text-independent). Experimental evidence suggests that training on multiple languages can generalize to unseen varieties while maintaining performance on seen varieties. We also found that it can reduce computational requirements for training models by an order of magnitude. Furthermore, during model inference on English data, we observe that leveraging a triage framework can reduce the number of calls to the more computationally expensive text-independent system by 73% (and reduce latency by 59%) while maintaining an EER no worse than the text-independent setup.

Even state-of-the-art speaker diarization systems exhibit high variance in error rates across different datasets, representing numerous use cases and domains. Furthermore, comparing across systems requires careful application of best practices such as dataset splits and metric definitions to allow for apples-to-apples comparison. We propose SDBench (Speaker Diarization Benchmark), an open-source benchmark suite that integrates 13 diverse datasets with built-in tooling for consistent and fine-grained analysis of speaker diarization performance for various on-device and server-side systems. SDBench enables reproducible evaluation and easy integration of new systems over time. To demonstrate the efficacy of SDBench, we built SpeakerKit, an inference efficiency-focused system built on top of Pyannote v3. SDBench enabled rapid execution of ablation studies that led to SpeakerKit being 9.6x faster than Pyannote v3 while achieving comparable error rates. We benchmark 6 state-of-the-art systems including Deepgram, AWS Transcribe, and Pyannote AI API, revealing important trade-offs between accuracy and speed.

Large Audio-Language Models (LALMs) are increasingly integrated into daily applications, yet their generative biases remain underexplored. Existing speech fairness benchmarks rely on synthetic speech and Multiple-Choice Questions (MCQs), both offering a fragmented view of fairness. We propose VIBE, a framework that evaluates generative bias through open-ended tasks such as personalized recommendations, using human-recorded speech. Unlike MCQs, our method allows stereotypical associations to manifest organically without predefined options, making it easily extensible to new tasks. Evaluating 12 state-of-the-art LALMs reveals systematic biases in realistic scenarios. Both gender and accent cues trigger statistically significant distributional shifts, and bias magnitude is strongly task-dependent.

Traditional voice conversion (VC) methods typically attempt to separate speaker identity and linguistic information into distinct representations, which are then combined to reconstruct the audio. However, effectively disentangling these factors remains challenging, often leading to information loss during training. In this paper, we propose a new approach that leverages synthetic speech data generated by a high-quality, pretrained multispeaker text-to-speech (TTS) model. Specifically, synthetic data pairs that share the same linguistic content but differ in speaker identity are used as input-output pairs to train the voice conversion model. This enables the model to learn a direct mapping between source and target voices, effectively capturing speaker-specific characteristics while preserving linguistic content. Additionally, we introduce a flexible training strategy for any-to-any voice conversion that generalizes well to unseen speakers and new languages, enhancing adaptability and performance in zero-shot scenarios. Our experiments show that our proposed method achieves a 16.35% relative reduction in word error rate and a 5.91% improvement in speaker cosine similarity, outperforming several state-of-the-art methods. Voice conversion samples can be accessed at: https://oovc-emnlp-2025.github.io/

As expressive text-to-speech (TTS) and voice conversion (VC) systems increasingly generate non-verbal vocalizations (NVVs) to enhance naturalness, reliable speaker verification (SV) becomes essential to objectively assess identity consistency across both verbal and non-verbal segments. Yet current SV systems generalize poorly to NVVs, and fine-tuning on NVV data causes catastrophic forgetting of speech performance. We present the first systematic study across 10 NVV types and propose a framework combining frozen Data2Vec self-supervised features with ECAPA-TDNN, enhanced by a Mixture of Experts (MoE) module with learned domain-aware routing. A conditional distillation loss on speech inputs via a pretrained teacher retains speech-to-speech accuracy, while a contrastive loss bridges the speech-NVV domain gap. Our method reduces speech-NVV EER from 38.93% to 22.66% over a pretrained baseline, and improves speech EER from 13.17% to 9.24% via distillation.

Self-supervised pre-training recently demonstrates success on large-scale multimodal data, and state-of-the-art contrastive learning methods often enforce the feature consistency from cross-modality inputs, such as video/audio or video/text pairs. Despite its convenience to formulate and leverage in practice, such cross-modality alignment (CMA) is only a weak and noisy supervision, since two modalities can be semantically misaligned even they are temporally aligned. For example, even in the commonly adopted instructional videos, a speaker can sometimes refer to something that is not visually present in the current frame; and the semantic misalignment would only be more unpredictable for the raw videos from the internet. We conjecture that might cause conflicts and biases among modalities, and may hence prohibit CMA from scaling up to training with larger and more heterogeneous data. This paper first verifies our conjecture by observing that, even in the latest VATT pre-training using only instructional videos, there exist strong gradient conflicts between different CMA losses within the same video, audio, text triplet, indicating them as the noisy source of supervision. We then propose to harmonize such gradients, via two techniques: (i) cross-modality gradient realignment: modifying different CMA loss gradients for each sample triplet, so that their gradient directions are more aligned; and (ii) gradient-based curriculum learning: leveraging the gradient conflict information on an indicator of sample noisiness, to develop a curriculum learning strategy to prioritize training on less noisy sample triplets. Applying those techniques to pre-training VATT on the HowTo100M dataset, we consistently improve its performance on different downstream tasks. Moreover, we are able to scale VATT pre-training to more complicated non-narrative Youtube8M dataset to further improve the state-of-the-arts.

Personalised speech enhancement (PSE), which extracts only the speech of a target user and removes everything else from a recorded audio clip, can potentially improve users' experiences of audio AI modules deployed in the wild. To support a large variety of downstream audio tasks, such as real-time ASR and audio-call enhancement, a PSE solution should operate in a streaming mode, i.e., input audio cleaning should happen in real-time with a small latency and real-time factor. Personalisation is typically achieved by extracting a target speaker's voice profile from an enrolment audio, in the form of a static embedding vector, and then using it to condition the output of a PSE model. However, a fixed target speaker embedding may not be optimal under all conditions. In this work, we present a streaming Transformer-based PSE model and propose a novel cross-attention approach that gives adaptive target speaker representations. We present extensive experiments and show that our proposed cross-attention approach outperforms competitive baselines consistently, even when our model is only approximately half the size.

We introduce LRLspoof, a large-scale multilingual synthetic-speech corpus for cross-lingual spoof detection, comprising 2,732 hours of audio generated with 24 open-source TTS systems across 66 languages, including 45 low-resource languages under our operational definition. To evaluate robustness without requiring target-domain bonafide speech, we benchmark 11 publicly available countermeasures using threshold transfer: for each model we calibrate an EER operating point on pooled external benchmarks and apply the resulting threshold, reporting spoof rejection rate (SRR). Results show model-dependent cross-lingual disparity, with spoof rejection varying markedly across languages even under controlled conditions, highlighting language as an independent source of domain shift in spoof detection. The dataset is publicly available at https://huggingface.co/datasets/MTUCI/LRLspoof{\underline{\textit{HuggingFace}}} and https://modelscope.cn/datasets/lab260/LRLspoof{\underline{\textit{ModelScope}}}

Several recent end-to-end text-to-speech (TTS) models enabling single-stage training and parallel sampling have been proposed, but their sample quality does not match that of two-stage TTS systems. In this work, we present a parallel end-to-end TTS method that generates more natural sounding audio than current two-stage models. Our method adopts variational inference augmented with normalizing flows and an adversarial training process, which improves the expressive power of generative modeling. We also propose a stochastic duration predictor to synthesize speech with diverse rhythms from input text. With the uncertainty modeling over latent variables and the stochastic duration predictor, our method expresses the natural one-to-many relationship in which a text input can be spoken in multiple ways with different pitches and rhythms. A subjective human evaluation (mean opinion score, or MOS) on the LJ Speech, a single speaker dataset, shows that our method outperforms the best publicly available TTS systems and achieves a MOS comparable to ground truth.

This paper proposes attentive statistics pooling for deep speaker embedding in text-independent speaker verification. In conventional speaker embedding, frame-level features are averaged over all the frames of a single utterance to form an utterance-level feature. Our method utilizes an attention mechanism to give different weights to different frames and generates not only weighted means but also weighted standard deviations. In this way, it can capture long-term variations in speaker characteristics more effectively. An evaluation on the NIST SRE 2012 and the VoxCeleb data sets shows that it reduces equal error rates (EERs) from the conventional method by 7.5% and 8.1%, respectively.

This paper presents our systems (denoted as T13) for the singing voice conversion challenge (SVCC) 2023. For both in-domain and cross-domain English singing voice conversion (SVC) tasks (Task 1 and Task 2), we adopt a recognition-synthesis approach with self-supervised learning-based representation. To achieve data-efficient SVC with a limited amount of target singer/speaker's data (150 to 160 utterances for SVCC 2023), we first train a diffusion-based any-to-any voice conversion model using publicly available large-scale 750 hours of speech and singing data. Then, we finetune the model for each target singer/speaker of Task 1 and Task 2. Large-scale listening tests conducted by SVCC 2023 show that our T13 system achieves competitive naturalness and speaker similarity for the harder cross-domain SVC (Task 2), which implies the generalization ability of our proposed method. Our objective evaluation results show that using large datasets is particularly beneficial for cross-domain SVC.

Speech transcripts used as long-term memory must preserve both words and stable speaker identities. Existing meeting-transcription metrics either ignore speakers or remap anonymous speakers independently in each recording, so they cannot measure whether the same person retains one identity across meetings. We evaluate persistent speaker attribution with Speaker Identified cpWER (SI-cpWER), which scores a corpus under one global speaker-ID assignment. The benchmark covers five commercial diarize-then-identify cascades, two open academic baselines, and ThyVoice on the full 129-meeting CHiME-8 NOTSOFAR evaluation set in clean and noiseaugmented form, plus CHiME-6. ThyVoice is our end-to-end reference system; it repairs overlap and gates the evidence used to create and update voiceprints. Requiring persistent identity changes the commercial ranking: ThyVoice records lower SI-cpWER than every evaluated commercial cascade in all three conditions and the lowest mean in the full panel, 47.13 versus 54.75 for the next system. Complementary lexical, diarization, per-recording attribution, and speaker-clustering diagnostics characterize upstream error surfaces in the final attributed record. These results show why persistent attribution must be evaluated directly in systems that reuse conversations across time.

Text-output scores alone do not show whether quantization preserves performance on speech tasks whose target labels cannot be recovered from the transcript. We evaluate fixed mixed 4/8-bit Qwen2-Audio-7B-Instruct allocations averaging 6 and 7 bits per parameter on 508 English-to-German FLEURS utterances and on 512 RAVDESS emotion clips from 16 speakers. The BLEU and chrF differences from half precision (FP16) have intervals that include zero for both allocations. On RAVDESS, the same two sentences occur equally often with every emotion label. The absolute accuracy differences from FP16 are -3.71% for 6 bit and -1.17% for 7 bit. The 6-bit speaker interval excludes zero and an exact two-sided sign-flip test gives p=0.0148; the 7-bit interval includes zero. Same-budget controls do not identify either selected allocation as best. This case study shows why translation scores and performance on tasks beyond the transcript need separate evaluation.

Large language models are increasingly used as decision aids whose probability judgments shape downstream choices. Whether those judgments carry a systematic directional tilt has been hard to detect: calibration metrics aggregate unsigned errors, and naturalistic uncertainty offers no ground-truth probability. When an LLM rates a startup's success at 70% but its failure at 15%, the missing 15 points expose a distortion no aggregate score flags. We introduce OptimismBench, which detects directional bias with inverted pairs: each scenario elicits both P(success) and P(failure), and asymmetry between the two framings yields a signed bias score without ground truth. Across 16 models from 8 providers, fourteen are optimistic; pessimism appears only in Anthropic's frontier tier. Eleven matched base-versus-chat pairs across four families show post-training sets the sign of the bias, with opposite shifts in different families. The pattern survives prompt, temperature, perspective, and self-debiasing ablations. A seventeen-model six-language comparison further shows model identity dominates language, with inter-model variance at 4.7x inter-language variance. We release 3,870 items across 10 languages for per-model directional-bias auditing. When alignment makes a model more helpful, it also tilts its probabilities; downstream pipelines inherit the tilt by default.

Large audio-language models (LALMs) extend text-based LLMs with auditory understanding, offering new opportunities for multimodal applications. While their perception, reasoning, and task performance have been widely studied, their safety alignment under paralinguistic variation remains underexplored. This work systematically investigates the role of speaker emotion. We construct a dataset of malicious speech instructions expressed across multiple emotions and intensities, and evaluate several state-of-the-art LALMs. Our results reveal substantial safety inconsistencies: different emotions elicit varying levels of unsafe responses, and the effect of intensity is non-monotonic, with medium expressions often posing the greatest risk. These findings highlight an overlooked vulnerability in LALMs and call for alignment strategies explicitly designed to ensure robustness under emotional variation, a prerequisite for trustworthy deployment in real-world settings.

We introduce a sophisticated multi-speaker speech data simulator, specifically engineered to generate multi-speaker speech recordings. A notable feature of this simulator is its capacity to modulate the distribution of silence and overlap via the adjustment of statistical parameters. This capability offers a tailored training environment for developing neural models suited for speaker diarization and voice activity detection. The acquisition of substantial datasets for speaker diarization often presents a significant challenge, particularly in multi-speaker scenarios. Furthermore, the precise time stamp annotation of speech data is a critical factor for training both speaker diarization and voice activity detection. Our proposed multi-speaker simulator tackles these problems by generating large-scale audio mixtures that maintain statistical properties closely aligned with the input parameters. We demonstrate that the proposed multi-speaker simulator generates audio mixtures with statistical properties that closely align with the input parameters derived from real-world statistics. Additionally, we present the effectiveness of speaker diarization and voice activity detection models, which have been trained exclusively on the generated simulated datasets.

Cross-attention is a core mechanism in encoder-decoder architectures, widespread in many fields, including speech-to-text (S2T) processing. Its scores have been repurposed for various downstream applications--such as timestamp estimation and audio-text alignment--under the assumption that they reflect the dependencies between input speech representation and the generated text. While the explanatory nature of attention mechanisms has been widely debated in the broader NLP literature, this assumption remains largely unexplored within the speech domain. To address this gap, we assess the explanatory power of cross-attention in S2T models by comparing its scores to input saliency maps derived from feature attribution. Our analysis spans monolingual and multilingual, single-task and multi-task models at multiple scales, and shows that attention scores moderately to strongly align with saliency-based explanations, particularly when aggregated across heads and layers. However, it also shows that cross-attention captures only about 50% of the input relevance and, in the best case, only partially reflects how the decoder attends to the encoder's representations--accounting for just 52-75% of the saliency. These findings uncover fundamental limitations in interpreting cross-attention as an explanatory proxy, suggesting that it offers an informative yet incomplete view of the factors driving predictions in S2T models.

We propose a statistical framework built on latent variable modeling for scaling laws of large language models (LLMs). Our work is motivated by the rapid emergence of numerous new LLM families with distinct architectures and training strategies, evaluated on an increasing number of benchmarks. This heterogeneity makes a single global scaling curve inadequate for capturing how performance varies across families and benchmarks. To address this, we propose a latent variable modeling framework in which each LLM family is associated with a latent variable that captures the common underlying features in that family. An LLM's performance on different benchmarks is then driven by its latent skills, which are jointly determined by the latent variable and the model's own observable features. We develop an estimation procedure for this latent variable model and establish its statistical properties. We also design efficient numerical algorithms that support estimation and various downstream tasks. Empirically, we evaluate the approach on 12 widely used benchmarks from the Open LLM Leaderboard (v1/v2).

State-of-the-art speaker verification systems are inherently dependent on some kind of human supervision as they are trained on massive amounts of labeled data. However, manually annotating utterances is slow, expensive and not scalable to the amount of data available today. In this study, we explore self-supervised learning for speaker verification by learning representations directly from raw audio. The objective is to produce robust speaker embeddings that have small intra-speaker and large inter-speaker variance. Our approach is based on recent information maximization learning frameworks and an intensive data augmentation pre-processing step. We evaluate the ability of these methods to work without contrastive samples before showing that they achieve better performance when combined with a contrastive loss. Furthermore, we conduct experiments to show that our method reaches competitive results compared to existing techniques and can get better performances compared to a supervised baseline when fine-tuned with a small portion of labeled data.

Although high-fidelity speech can be obtained for intralingual speech synthesis, cross-lingual text-to-speech (CTTS) is still far from satisfactory as it is difficult to accurately retain the speaker timbres(i.e. speaker similarity) and eliminate the accents from their first language(i.e. nativeness). In this paper, we demonstrated that vector-quantized(VQ) acoustic feature contains less speaker information than mel-spectrogram. Based on this finding, we propose a novel dual speaker embedding TTS (DSE-TTS) framework for CTTS with authentic speaking style. Here, one embedding is fed to the acoustic model to learn the linguistic speaking style, while the other one is integrated into the vocoder to mimic the target speaker's timbre. Experiments show that by combining both embeddings, DSE-TTS significantly outperforms the state-of-the-art SANE-TTS in cross-lingual synthesis, especially in terms of nativeness.

Automatic speech recognition (ASR) systems have achieved near-human accuracy on curated benchmarks, yet still fail in real-world voice agents under conditions that current evaluations do not systematically cover. Without diagnostic tools that isolate specific failure factors, practitioners cannot anticipate which conditions, in which languages, will cause what degree of degradation. We introduce WildASR, a multilingual (four-language) diagnostic benchmark sourced entirely from real human speech that factorizes ASR robustness along three axes: environmental degradation, demographic shift, and linguistic diversity. Evaluating seven widely used ASR systems, we find severe and uneven performance degradation, and model robustness does not transfer across languages or conditions. Critically, models often hallucinate plausible but unspoken content under partial or degraded inputs, creating concrete safety risks for downstream agent behavior. Our results demonstrate that targeted, factor-isolated evaluation is essential for understanding and improving ASR reliability in production systems. Besides the benchmark itself, we also present three analytical tools that practitioners can use to guide deployment decisions.

Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) exhibit fundamentally different behaviors in enhancing multi-task reasoning for large language models (LLMs). Our preliminary experiments revealed a phenomenon: SFT suffers from severe task conflicts under multi-stage training, whereas RL enables stable coexistence across diverse tasks. Empirically, we trace this to the parameter level, observing that RL induces sparse and approximately orthogonal updates across tasks. We provide a theoretical explanation for this mechanism by analyzing multi-task gradient interference. Our results reveal a distinction: interference in SFT is norm-limited, scaling with the absolute gradient magnitude, whereas interference in RL is variance-limited, bounded by the gradient variance induced by advantage normalization and on-policy optimization. This small variance bound yields near-orthogonal optimization directions across tasks. Leveraging this insight, we propose Parallel-RL, a paradigm that decouples multi-task training, significantly improving efficiency and flexibility.

Large Language Models (LLMs) are predominantly evaluated on Standard American English (SAE), often overlooking the diversity of global English varieties. This narrow focus may raise fairness concerns as degraded performance on non-standard varieties can lead to unequal benefits for users worldwide. Therefore, it is critical to extensively evaluate the linguistic robustness of LLMs on multiple non-standard English varieties. We introduce Trans-EnV, a framework that automatically transforms SAE datasets into multiple English varieties to evaluate the linguistic robustness. Our framework combines (1) linguistics expert knowledge to curate variety-specific features and transformation guidelines from linguistic literature and corpora, and (2) LLM-based transformations to ensure both linguistic validity and scalability. Using Trans-EnV, we transform six benchmark datasets into 38 English varieties and evaluate seven state-of-the-art LLMs. Our results reveal significant performance disparities, with accuracy decreasing by up to 46.3% on non-standard varieties. These findings highlight the importance of comprehensive linguistic robustness evaluation across diverse English varieties. Each construction of Trans-EnV was validated through rigorous statistical testing and consultation with a researcher in the field of second language acquisition, ensuring its linguistic validity. Our code and datasets are publicly available at https://github.com/jiyounglee-0523/TransEnV and https://huggingface.co/collections/jiyounglee0523/transenv-681eadb3c0c8cf363b363fb1.

Multimodal foundation models are prone to hallucination, generating outputs that either contradict the input or are not grounded by factual information. Given the diversity in architectures, training data and instruction tuning techniques, there can be large variations in systems' susceptibility to hallucinations. To assess system hallucination robustness, hallucination ranking approaches have been developed for specific tasks such as image captioning, question answering, summarization, or biography generation. However, these approaches typically compare model outputs to gold-standard references or labels, limiting hallucination benchmarking for new domains. This work proposes "CrossCheckGPT", a reference-free universal hallucination ranking for multimodal foundation models. The core idea of CrossCheckGPT is that the same hallucinated content is unlikely to be generated by different independent systems, hence cross-system consistency can provide meaningful and accurate hallucination assessment scores. CrossCheckGPT can be applied to any model or task, provided that the information consistency between outputs can be measured through an appropriate distance metric. Focusing on multimodal large language models that generate text, we explore two information consistency measures: CrossCheck-explicit and CrossCheck-implicit. We showcase the applicability of our method for hallucination ranking across various modalities, namely the text, image, and audio-visual domains. Further, we propose the first audio-visual hallucination benchmark, "AVHalluBench", and illustrate the effectiveness of CrossCheckGPT, achieving correlations of 98% and 89% with human judgements on MHaluBench and AVHalluBench, respectively.

This work presents the first systematic investigation of speech bias in multilingual MLLMs. We construct and release the BiasInEar dataset, a speech-augmented benchmark based on Global MMLU Lite, spanning English, Chinese, and Korean, balanced by gender and accent, and totaling 70.8 hours (approx4,249 minutes) of speech with 11,200 questions. Using four complementary metrics (accuracy, entropy, APES, and Fleiss' κ), we evaluate nine representative models under linguistic (language and accent), demographic (gender), and structural (option order) perturbations. Our findings reveal that MLLMs are relatively robust to demographic factors but highly sensitive to language and option order, suggesting that speech can amplify existing structural biases. Moreover, architectural design and reasoning strategy substantially affect robustness across languages. Overall, this study establishes a unified framework for assessing fairness and robustness in speech-integrated LLMs, bridging the gap between text- and speech-based evaluation. The resources can be found at https://github.com/ntunlplab/BiasInEar.

We introduce an approach to multilingual speech synthesis which uses the meta-learning concept of contextual parameter generation and produces natural-sounding multilingual speech using more languages and less training data than previous approaches. Our model is based on Tacotron 2 with a fully convolutional input text encoder whose weights are predicted by a separate parameter generator network. To boost voice cloning, the model uses an adversarial speaker classifier with a gradient reversal layer that removes speaker-specific information from the encoder. We arranged two experiments to compare our model with baselines using various levels of cross-lingual parameter sharing, in order to evaluate: (1) stability and performance when training on low amounts of data, (2) pronunciation accuracy and voice quality of code-switching synthesis. For training, we used the CSS10 dataset and our new small dataset based on Common Voice recordings in five languages. Our model is shown to effectively share information across languages and according to a subjective evaluation test, it produces more natural and accurate code-switching speech than the baselines.

Target-speaker speech processing (TS) tasks, such as target-speaker automatic speech recognition (TS-ASR), target speech extraction (TSE), and personal voice activity detection (p-VAD), are important for extracting information about a desired speaker's speech even when it is corrupted by interfering speakers. While most studies have focused on training schemes or system architectures for each specific task, the auxiliary network for embedding target-speaker cues has not been investigated comprehensively in a unified cross-task evaluation. Therefore, this paper aims to address a fundamental question: what is the preferred speaker embedding for TS tasks? To this end, for the TS-ASR, TSE, and p-VAD tasks, we compare pre-trained speaker encoders (i.e., self-supervised or speaker recognition models) that compute speaker embeddings from pre-recorded enrollment speech of the target speaker with ideal speaker embeddings derived directly from the target speaker's identity in the form of a one-hot vector. To further understand the properties of ideal speaker embedding, we optimize it using a gradient-based approach to improve performance on the TS task. Our analysis reveals that speaker verification performance is somewhat unrelated to TS task performances, the one-hot vector outperforms enrollment-based ones, and the optimal embedding depends on the input mixture.

Large language models (LLMs) are rapidly transforming social science research by enabling the automation of labor-intensive tasks like data annotation and text analysis. However, LLM outputs vary significantly depending on the implementation choices made by researchers (e.g., model selection, prompting strategy, or temperature settings). Such variation can introduce systematic biases and random errors, which propagate to downstream analyses and cause Type I, Type II, Type S, or Type M errors. We call this LLM hacking. We quantify the risk of LLM hacking by replicating 37 data annotation tasks from 21 published social science research studies with 18 different models. Analyzing 13 million LLM labels, we test 2,361 realistic hypotheses to measure how plausible researcher choices affect statistical conclusions. We find incorrect conclusions based on LLM-annotated data in approximately one in three hypotheses for state-of-the-art models, and in half the hypotheses for small language models. While our findings show that higher task performance and better general model capabilities reduce LLM hacking risk, even highly accurate models do not completely eliminate it. The risk of LLM hacking decreases as effect sizes increase, indicating the need for more rigorous verification of findings near significance thresholds. Our extensive analysis of LLM hacking mitigation techniques emphasizes the importance of human annotations in reducing false positive findings and improving model selection. Surprisingly, common regression estimator correction techniques are largely ineffective in reducing LLM hacking risk, as they heavily trade off Type I vs. Type II errors. Beyond accidental errors, we find that intentional LLM hacking is unacceptably simple. With few LLMs and just a handful of prompt paraphrases, anything can be presented as statistically significant.

We introduce a new dataset of conversational speech representing English from India, Nigeria, and the United States. The Multi-Dialect Dataset of Dialogues (MD3) strikes a new balance between open-ended conversational speech and task-oriented dialogue by prompting participants to perform a series of short information-sharing tasks. This facilitates quantitative cross-dialectal comparison, while avoiding the imposition of a restrictive task structure that might inhibit the expression of dialect features. Preliminary analysis of the dataset reveals significant differences in syntax and in the use of discourse markers. The dataset, which will be made publicly available with the publication of this paper, includes more than 20 hours of audio and more than 200,000 orthographically-transcribed tokens.

The rapid advancement of large language models (LLMs) has accelerated the development of multi-modal models capable of vocal communication. Unlike text-based interactions, speech conveys rich and diverse information, including semantic content, acoustic variations, paralanguage cues, and environmental context. However, existing evaluations of speech interaction models predominantly focus on the quality of their textual responses, often overlooking critical aspects of vocal performance and lacking benchmarks with vocal-specific test instances. To address this gap, we propose VocalBench, a comprehensive benchmark designed to evaluate speech interaction models' capabilities in vocal communication. VocalBench comprises 9,400 carefully curated instances across four key dimensions: semantic quality, acoustic performance, conversational abilities, and robustness. It covers 16 fundamental skills essential for effective vocal interaction. Experimental results reveal significant variability in current model capabilities, each exhibiting distinct strengths and weaknesses, and provide valuable insights to guide future research in speech-based interaction systems. Code and evaluation instances are available at https://github.com/SJTU-OmniAgent/VocalBench.

Phoneme-level models transcribe or generate speech as a sequence of phonemes, the smallest sound units that distinguish words. These models enable fine-grained pronunciation control and understanding, yet often fail on input that does not match any single training language, such as speech alternating between two languages, known as code-switching, or low-resource languages absent from training. We identify a systematic failure mode behind this, phonological interference: models assume the input is in a single language and impose its phonology, overriding local phoneme-level decisions that conflict with the assumed language. We measure interference by how often a model retains phonemes that one language has but the other lacks. On code-switched input, two phone recognizers (speech-to-phoneme models) and a phoneme-conditioned text-to-speech model lose 32% to 79% of these phonemes, but lose far fewer of the phonemes both languages share. On unseen languages, we find that phone recognizers impose the phonology of the training language they assign to the speech, and the more confident the assignment, the more they lose phonemes the unseen language has but the assigned language lacks. We probe the models' language estimate from their internal activations, and trace interference to a low dimensional subspace. On monolingual speech, steering this subspace toward another language makes the model lose the phonemes that only the original language uses and produce phonemes that only the target language has. We introduce windowed language estimation (WLE), an inference time repair that replaces the model's language estimate in this subspace with one computed from a short window around each position. On code-switched input, WLE removes 34% to 69% of the interference in all three models, and in the recognizers it leaves monolingual performance essentially unchanged.

We study device-addressed speech detection under pre-ASR edge deployment constraints, where systems must decide whether to forward audio before transcription under strict latency and compute limits. We show that, in multi-speaker environments with temporally ambiguous utterances, this task is more effectively modelled as a sequential routing problem over interaction history than as an utterance-local classification task. We formalize this as Sequential Device-Addressed Routing (SDAR) and present the Selective Attention System (SAS), an on-device implementation that instantiates this formulation. On a held-out 60-hour multi-speaker English test set, the primary audio-only configuration achieves F1=0.86 (precision=0.89, recall=0.83); with an optional camera, audio+video fusion raises F1 to 0.95 (precision=0.97, recall=0.93). Removing causal interaction history (Stage~3) reduced F1 from 0.95 to 0.57+/-0.03 in the audio+video configuration under our evaluation protocol. Among the tested components, this was the largest observed ablation effect, indicating that short-horizon interaction history carries substantial decision-relevant information in the evaluated setting. SAS runs fully on-device on ARM Cortex-A class hardware (<150 ms latency, <20 MB footprint). All results are from internal evaluation on a proprietary dataset evaluated primarily in English; a 5-hour evaluation subset may be shared for independent verification (Section 8.8).

Multilingual proficiency presents a significant challenge for large language models (LLMs). English-centric models are usually suboptimal in other languages, particularly those that are linguistically distant from English. This performance discrepancy mainly stems from the imbalanced distribution of training data across languages during pre-training and instruction tuning stages. To address this problem, we propose a novel approach called CrossIn, which utilizes a mixed composition of cross-lingual instruction tuning data. Our method leverages the compressed representation shared by various languages to efficiently enhance the model's task-solving capabilities and multilingual proficiency within a single process. In addition, we introduce a multi-task and multi-faceted benchmark to evaluate the effectiveness of CrossIn. Experimental results demonstrate that our method substantially improves performance across tasks and languages, and we provide extensive insights into the impact of cross-lingual data volume and the integration of translation data on enhancing multilingual consistency and accuracy.

Speech processing algorithms often rely on statistical knowledge of the underlying process. Despite many years of research, however, the debate on the most appropriate statistical model for speech still continues. Speech is commonly modeled as a wide-sense stationary (WSS) process. However, the use of the WSS model for spectrally correlated processes is fundamentally wrong, as WSS implies spectral uncorrelation. In this paper, we demonstrate that voiced speech can be more accurately represented as a cyclostationary (CS) process. By employing the CS rather than the WSS model for processes that are inherently correlated across frequency, it is possible to improve the estimation of cross-power spectral densities (PSDs), source separation, and beamforming. We illustrate how the correlation between harmonic frequencies of CS processes can enhance system identification, and validate our findings using both simulated and real speech data.

We explore cross-lingual multi-speaker speech synthesis and cross-lingual voice conversion applied to data augmentation for automatic speech recognition (ASR) systems. Through extensive experiments, we show that our approach permits the application of speech synthesis and voice conversion to improve ASR systems on a target language using only one target-language speaker during model training. We managed to close the gap between ASR models trained with synthesized versus human speech compared to other works that use many speakers. Finally, we show that it is possible to obtain promising ASR training results with our data augmentation method using only a single real speaker in a target language.

Personalizing automatic speech recognition (ASR) systems for non-normative speech, such as dysarthric and aphasic speech, is challenging. While speaker-specific fine-tuning (SS-FT) is widely used, it is typically initialized directly from a generic pre-trained model. Whether speaker-independent adaptation provides a stronger initialization prior under such mismatch remains unclear. In this work, we propose a two-stage adaptation framework consisting of speaker-independent fine-tuning (SI-FT) on multi-speaker non-normative data followed by SS-FT, and evaluate it through a controlled comparison with direct SS-FT under identical per-speaker conditions. Experiments on AphasiaBank and UA-Speech with Whisper-Large-v3 and Qwen3-ASR, alongside evaluation on typical-speech datasets TED-LIUM v3 and FLEURS, show that two-stage adaptation consistently improves personalization while maintaining manageable out-of-domain (OOD) trade-offs.