A Study of the Removability of Speaker-Adversarial Perturbations

Recent advancements in adversarial attacks have demonstrated their effectiveness in misleading speaker recognition models, making wrong predictions about speaker identities. On the other hand, defense techniques against speaker-adversarial attacks focus on reducing the effects of speaker-adversarial perturbations on speaker attribute extraction. These techniques do not seek to fully remove the perturbations and restore the original speech. To this end, this paper studies the removability of speaker-adversarial perturbations. Specifically, the investigation is conducted assuming various degrees of awareness of the perturbation generator across three scenarios: ignorant, semi-informed, and well-informed. Besides, we consider both the optimization-based and feedforward perturbation generation methods. Experiments conducted on the LibriSpeech dataset demonstrated that: 1) in the ignorant scenario, speaker-adversarial perturbations cannot be eliminated, although their impact on speaker attribute extraction is reduced, 2) in the semi-informed scenario, the speaker-adversarial perturbations cannot be fully removed, while those generated by the feedforward model can be considerably reduced, and 3) in the well-informed scenario, speaker-adversarial perturbations are nearly eliminated, allowing for the restoration of the original speech. Audio samples can be found in https://voiceprivacy.github.io/Perturbation-Generation-Removal/.

Are Multimodal Foundation Models All That Is Needed for Emofake Detection?

In this work, we investigate multimodal foundation models (MFMs) for EmoFake detection (EFD) and hypothesize that they will outperform audio foundation models (AFMs). MFMs due to their cross-modal pre-training, learns emotional patterns from multiple modalities, while AFMs rely only on audio. As such, MFMs can better recognize unnatural emotional shifts and inconsistencies in manipulated audio, making them more effective at distinguishing real from fake emotional expressions. To validate our hypothesis, we conduct a comprehensive comparative analysis of state-of-the-art (SOTA) MFMs (e.g. LanguageBind) alongside AFMs (e.g. WavLM). Our experiments confirm that MFMs surpass AFMs for EFD. Beyond individual foundation models (FMs) performance, we explore FMs fusion, motivated by findings in related research areas such synthetic speech detection and speech emotion recognition. To this end, we propose SCAR, a novel framework for effective fusion. SCAR introduces a nested cross-attention mechanism, where representations from FMs interact at two stages sequentially to refine information exchange. Additionally, a self-attention refinement module further enhances feature representations by reinforcing important cross-FM cues while suppressing noise. Through SCAR with synergistic fusion of MFMs, we achieve SOTA performance, surpassing both standalone FMs and conventional fusion approaches and previous works on EFD.

Benchmarking Large Pretrained Multilingual Models on Québec French Speech Recognition

We evaluate the performance of large pretrained multilingual speech recognition models on a regional variety of French spoken in Qu\'ebec, Canada, in terms of speed, word error rate and semantic accuracy. To this end we build a benchmark and evaluation pipeline based on the CommissionsQc datasets, a corpus of spontaneous conversations recorded during public inquiries recently held in Qu\'ebec. Published results for these models on well-known benchmarks such as FLEURS or CommonVoice are not good predictors of the performance we observe on CommissionsQC. Our results should be of interest for practitioners interested in building speech applications for realistic conditions or regional language varieties.

CS3-Bench: Evaluating and Enhancing Speech-to-Speech LLMs for Mandarin-English Code-Switching

The advancement of multimodal large language models has accelerated the development of speech-to-speech interaction systems. While natural monolingual interaction has been achieved, we find existing models exhibit deficiencies in language alignment. In our proposed Code-Switching Speech-to-Speech Benchmark (CS3-Bench), experiments on 7 mainstream models demonstrate a relative performance drop of up to 66% in knowledge-intensive question answering and varying degrees of misunderstanding in open-ended conversations. Starting from a model with severe performance deterioration, we propose both data constructions and training approaches to improve the language alignment capabilities, specifically employing Chain of Recognition (CoR) to enhance understanding and Keyword Highlighting (KH) to guide generation. Our approach improves the knowledge accuracy from 25.14% to 46.13%, with open-ended understanding rate from 64.5% to 86.5%, and significantly reduces pronunciation errors in the secondary language. CS3-Bench is available at https://huggingface.co/datasets/VocalNet/CS3-Bench.

Cross-Learning Fine-Tuning Strategy for Dysarthric Speech Recognition Via CDSD database

Dysarthric speech recognition faces challenges from severity variations and disparities relative to normal speech. Conventional approaches individually fine-tune ASR models pre-trained on normal speech per patient to prevent feature conflicts. Counter-intuitively, experiments reveal that multi-speaker fine-tuning (simultaneously on multiple dysarthric speakers) improves recognition of individual speech patterns. This strategy enhances generalization via broader pathological feature learning, mitigates speaker-specific overfitting, reduces per-patient data dependence, and improves target-speaker accuracy - achieving up to 13.15% lower WER versus single-speaker fine-tuning.

OLMoASR: Open Models and Data for Training Robust Speech Recognition Models

Improvements in training data scale and quality have led to significant advances, yet its influence in speech recognition remains underexplored. In this paper, we present a large-scale dataset, OLMoASR-Pool, and series of models, OLMoASR, to study and develop robust zero-shot speech recognition models. Beginning from OLMoASR-Pool, a collection of 3M hours of English audio and 17M transcripts, we design text heuristic filters to remove low-quality or mistranscribed data. Our curation pipeline produces a new dataset containing 1M hours of high-quality audio-transcript pairs, which we call OLMoASR-Mix. We use OLMoASR-Mix to train the OLMoASR-Mix suite of models, ranging from 39M (tiny.en) to 1.5B (large.en) parameters. Across all model scales, OLMoASR achieves comparable average performance to OpenAI's Whisper on short and long-form speech recognition benchmarks. Notably, OLMoASR-medium.en attains a 12.8\% and 11.0\% word error rate (WER) that is on par with Whisper's largest English-only model Whisper-medium.en's 12.4\% and 10.5\% WER for short and long-form recognition respectively (at equivalent parameter count). OLMoASR-Pool, OLMoASR models, and filtering, training and evaluation code will be made publicly available to further research on robust speech processing.

Streaming Sequence-to-Sequence Learning with Delayed Streams Modeling

We introduce Delayed Streams Modeling (DSM), a flexible formulation for streaming, multimodal sequence-to-sequence learning. Sequence-to-sequence generation is often cast in an offline manner, where the model consumes the complete input sequence before generating the first output timestep. Alternatively, streaming sequence-to-sequence rely on learning a policy for choosing when to advance on the input stream, or write to the output stream. DSM instead models already time-aligned streams with a decoder-only language model. By moving the alignment to a pre-processing step,and introducing appropriate delays between streams, DSM provides streaming inference of arbitrary output sequences, from any input combination, making it applicable to many sequence-to-sequence problems. In particular, given text and audio streams, automatic speech recognition (ASR) corresponds to the text stream being delayed, while the opposite gives a text-to-speech (TTS) model. We perform extensive experiments for these two major sequence-to-sequence tasks, showing that DSM provides state-of-the-art performance and latency while supporting arbitrary long sequences, being even competitive with offline baselines. Code, samples and demos are available at https://github.com/kyutai-labs/delayed-streams-modeling

A Bottom-up Framework with Language-universal Speech Attribute Modeling for Syllable-based ASR

We propose a bottom-up framework for automatic speech recognition (ASR) in syllable-based languages by unifying language-universal articulatory attribute modeling with syllable-level prediction. The system first recognizes sequences or lattices of articulatory attributes that serve as a language-universal, interpretable representation of pronunciation, and then transforms them into syllables through a structured knowledge integration process. We introduce two evaluation metrics, namely Pronunciation Error Rate (PrER) and Syllable Homonym Error Rate (SHER), to evaluate the model's ability to capture pronunciation and handle syllable ambiguities. Experimental results on the AISHELL-1 Mandarin corpus demonstrate that the proposed bottom-up framework achieves competitive performance and exhibits better robustness under low-resource conditions compared to the direct syllable prediction model. Furthermore, we investigate the zero-shot cross-lingual transferability on Japanese and demonstrate significant improvements over character- and phoneme-based baselines by 40% error rate reduction.

CAMÕES: A Comprehensive Automatic Speech Recognition Benchmark for European Portuguese

Existing resources for Automatic Speech Recognition in Portuguese are mostly focused on Brazilian Portuguese, leaving European Portuguese (EP) and other varieties under-explored. To bridge this gap, we introduce CAM\~OES, the first open framework for EP and other Portuguese varieties. It consists of (1) a comprehensive evaluation benchmark, including 46h of EP test data spanning multiple domains; and (2) a collection of state-of-the-art models. For the latter, we consider multiple foundation models, evaluating their zero-shot and fine-tuned performances, as well as E-Branchformer models trained from scratch. A curated set of 425h of EP was used for both fine-tuning and training. Our results show comparable performance for EP between fine-tuned foundation models and the E-Branchformer. Furthermore, the best-performing models achieve relative improvements above 35% WER, compared to the strongest zero-shot foundation model, establishing a new state-of-the-art for EP and other varieties.

LatPhon: Lightweight Multilingual G2P for Romance Languages and English

Grapheme-to-phoneme (G2P) conversion is a key front-end for text-to-speech (TTS), automatic speech recognition (ASR), speech-to-speech translation (S2ST) and alignment systems, especially across multiple Latin-script languages.We present LatPhon, a 7.5 M - parameter Transformer jointly trained on six such languages--English, Spanish, French, Italian, Portuguese, and Romanian. On the public ipa-dict corpus, it attains a mean phoneme error rate (PER) of 3.5%, outperforming the byte-level ByT5 baseline (5.4%) and approaching language-specific WFSTs (3.2%) while occupying 30 MB of memory, which makes on-device deployment feasible when needed. These results indicate that compact multilingual G2P can serve as a universal front-end for Latin-language speech pipelines.