Solange Rossato

Titouan Parcollet, Ha Nguyen, Solene Evain et al.

Self-supervised learning (SSL) is at the origin of unprecedented improvements in many different domains including computer vision and natural language processing. Speech processing drastically benefitted from SSL as most of the current domain-related tasks are now being approached with pre-trained models. This work introduces LeBenchmark 2.0 an open-source framework for assessing and building SSL-equipped French speech technologies. It includes documented, large-scale and heterogeneous corpora with up to 14,000 hours of heterogeneous speech, ten pre-trained SSL wav2vec 2.0 models containing from 26 million to one billion learnable parameters shared with the community, and an evaluation protocol made of six downstream tasks to complement existing benchmarks. LeBenchmark 2.0 also presents unique perspectives on pre-trained SSL models for speech with the investigation of frozen versus fine-tuned downstream models, task-agnostic versus task-specific pre-trained models as well as a discussion on the carbon footprint of large-scale model training. Overall, the newly introduced models trained on 14,000 hours of French speech outperform multilingual and previous LeBenchmark SSL models across the benchmark but also required up to four times more energy for pre-training.

Phuong-Hang Le, Valentin Pelloin, Arnault Chatelain et al.

We release Pantagruel models, a new family of self-supervised encoder models for French text and speech. Instead of predicting modality-tailored targets such as textual tokens or speech units, Pantagruel learns contextualized target representations in the feature space, allowing modality-specific encoders to capture linguistic and acoustic regularities more effectively. Separate models are pre-trained on large-scale French corpora, including Wikipedia, OSCAR and CroissantLLM for text, together with MultilingualLibriSpeech, LeBenchmark, and INA-100k for speech. INA-100k is a newly introduced 100,000-hour corpus of French audio derived from the archives of the Institut National de l'Audiovisuel (INA), the national repository of French radio and television broadcasts, providing highly diverse audio data. We evaluate Pantagruel across a broad range of downstream tasks spanning both modalities, including those from the standard French benchmarks such as FLUE or LeBenchmark. Across these tasks, Pantagruel models show competitive or superior performance compared to strong French baselines such as CamemBERT, FlauBERT, and LeBenchmark2.0, while maintaining a shared architecture that can seamlessly handle either speech or text inputs. These results confirm the effectiveness of feature-space self-supervised objectives for French representation learning and highlight Pantagruel as a robust foundation for multimodal speech-text understanding.

Felix Herron, Maja Hjuler, Solange Rossato et al.

Speech encoder models are known to model members of some speaker groups (SGs) better than others. However, there has been little work in establishing why this occurs on a technological level. To our knowledge, we present the first layerwise fairness analysis of pretrained self-supervised speech encoder models (S3Ms), probing each embedding layer for speaker identification (SID) automatic speech recognition (ASR). We find S3Ms produce embeddings biased against certain SGs for both tasks, starting at the very first latent layers. Furthermore, we find opposite patterns of layerwise bias for SID vs ASR for all models in our study: SID bias is minimized in layers that minimize overall SID error; on the other hand, ASR bias is maximized in layers that minimize overall ASR error. The inverse bias/error relationship for ASR is unaffected when probing S3Ms that are finetuned for ASR, suggesting SG-level bias is established during pretraining and is difficult to remove.

Felix Herron, Ange Richard, François Portet et al.

Many studies have shown automatic speech processing (ASR) systems have unequal performance across speakergroups (SG's). However, the manner in which such studies arrive at this conclusion is inconsistent. To pave the wayfor more reliable results in future studies, we lay out best practices for benchmarking ASR fairness based on literaturefrom machine learning fairness, social sciences, and speech science. We first describe the importance of preciselythe fairness hypothesis being interrogated, and tailoring fairness metrics to apply specifically to said hypothesis.We then examine several benchmarks used to rate ASR systems on fairness and discuss how their results can bemisconstrued without assiduous oversight into the intersections between SG's. We find that evaluating fairnessbased on single heterogeneous SG's, such as they are defined in fairness benchmarks, can lead to misidentifyingwhich SG's are actually being mistreated by ASR systems. We advocate for as fine-grained an analysis as possibleof the intersectionality of as many demographic variables as are available in the metadata of fairness corpora in orderto tease out such spurious correlations

Mahault Garnerin, Solange Rossato, Laurent Besacier

With the rise of artificial intelligence (AI) and the growing use of deep-learning architectures, the question of ethics, transparency and fairness of AI systems has become a central concern within the research community. We address transparency and fairness in spoken language systems by proposing a study about gender representation in speech resources available through the Open Speech and Language Resource platform. We show that finding gender information in open source corpora is not straightforward and that gender balance depends on other corpus characteristics (elicited/non elicited speech, low/high resource language, speech task targeted). The paper ends with recommendations about metadata and gender information for researchers in order to assure better transparency of the speech systems built using such corpora.

Hippolyte Fournier, Sina Alisamir, Safaa Azzakhnini et al.

The integration of artificial intelligence (AI) into healthcare has advanced significantly, yet affect recognition remains a major challenge, particularly in AI-assisted interventions such as Computerized Cognitive Training (CCT). The THERADIA-WoZ corpus was developed to enable multimodal affect recognition in the context of AI-driven CCT, focusing on an older adult population. This study extends the corpus by introducing a dataset collected from young adults, allowing direct comparison of affect recognition models across age groups. Our objective was to assess whether multimodal models based on dimensions borrowed from appraisal theories outperform those based on categorical labels and to evaluate their generalisation power across age corpora. After comparing both corpora, models were trained and tested using within-corpus, cross-corpus, and mixed-corpus evaluation. Results revealed that appraisal dimensions consistently outperformed categorical labels across all conditions, demonstrating greater predictive accuracy and stability. Notably, categorical labels failed to generalise across age corpora, as performance dropped to chance levels in cross-corpus evaluation. In contrast, appraisal dimensions maintained predictive performance above chance, reinforcing their robustness for cross-age affect recognition. Furthermore, training on both corpora did not improve generalisation beyond within-corpus training. The findings support the theoretical and practical advantages of appraisal dimensions over categorical labels in affective computing. They also highlight the importance of multimodal fusion and deep learning representations for emotion modeling. To facilitate future research, we provide an API for researchers interested in time-continuous emotion prediction, offering valuable tools for behavioral sciences to enhance the measurement of emotional states in various experimental settings.

Felix Herron, Solange Rossato, Alexandre Allauzen et al.

Modern automatic speech recognition (ASR) systems have been observed to function better for certain speaker groups (SGs) than others, despite recent gains in overall performance. One potential impediment to progress towards fairer ASR is a more nuanced understanding of the types of modeling errors that speech encoder models make, and in particular the difference between the structure of embeddings for high-performance and low-performance SGs. This paper proposes a framework typifying two types of error that can occur in modeling phonemes in ASR systems: random error/high variance in phoneme embedding, vs systematic error/embedding bias. We find that training phoneme classification probes only on a single, typically disadvantaged SG, sometimes improves performance for that SG, which is evidence for the existence of SG-level bias in phoneme embeddings. On the other hand, we find that speakers and SGs with higher levels of phoneme variance are the same as those with worse phoneme prediction accuracy. We conclude that both types of error are present in phoneme embeddings and both are candidate causes for SG-level unfairness in ASR, though random error is likely a greater hindrance to fairness than systematic error. Furthermore, we find that finetuning encoder models using a fairness-enhancing algorithm (domain enhancing and adversarial training) changes neither the benefits of in-domain phoneme classification probe training, nor measured levels of random embedding error.

Solene Evain, Ha Nguyen, Hang Le et al.

Self-Supervised Learning (SSL) using huge unlabeled data has been successfully explored for image and natural language processing. Recent works also investigated SSL from speech. They were notably successful to improve performance on downstream tasks such as automatic speech recognition (ASR). While these works suggest it is possible to reduce dependence on labeled data for building efficient speech systems, their evaluation was mostly made on ASR and using multiple and heterogeneous experimental settings (most of them for English). This questions the objective comparison of SSL approaches and the evaluation of their impact on building speech systems. In this paper, we propose LeBenchmark: a reproducible framework for assessing SSL from speech. It not only includes ASR (high and low resource) tasks but also spoken language understanding, speech translation and emotion recognition. We also focus on speech technologies in a language different than English: French. SSL models of different sizes are trained from carefully sourced and documented datasets. Experiments show that SSL is beneficial for most but not all tasks which confirms the need for exhaustive and reliable benchmarks to evaluate its real impact. LeBenchmark is shared with the scientific community for reproducible research in SSL from speech.

Mahault Garnerin, Solange Rossato, Laurent Besacier

This paper analyzes the gender representation in four major corpora of French broadcast. These corpora being widely used within the speech processing community, they are a primary material for training automatic speech recognition (ASR) systems. As gender bias has been highlighted in numerous natural language processing (NLP) applications, we study the impact of the gender imbalance in TV and radio broadcast on the performance of an ASR system. This analysis shows that women are under-represented in our data in terms of speakers and speech turns. We introduce the notion of speaker role to refine our analysis and find that women are even fewer within the Anchor category corresponding to prominent speakers. The disparity of available data for both gender causes performance to decrease on women. However this global trend can be counterbalanced for speaker who are used to speak in the media when sufficient amount of data is available.