Théo Lepage
Ph.D. student in Machine Learning → learning robust representations for speaker & language recognition
Paris, Île-de-France, France
637 abonnés
+ de 500 relations
Activité
-
✨ Nous sommes ravis d'annoncer la vente de notre drone compact SEEKER à Groupe Coopératif Vivadour ! ✨ Récemment, Léa Le Gall, de notre service…
✨ Nous sommes ravis d'annoncer la vente de notre drone compact SEEKER à Groupe Coopératif Vivadour ! ✨ Récemment, Léa Le Gall, de notre service…
Aimé par Théo Lepage
-
I had the opportunity to present our work on hologram verification using weakly supervised learning at ICDAR 2024 as a poster presentation. The paper…
I had the opportunity to present our work on hologram verification using weakly supervised learning at ICDAR 2024 as a poster presentation. The paper…
Aimé par Théo Lepage
-
Four (4 !) pyannoteAI speaker diarization models among the 10 most downloaded audio models on Hugging Face 🎉 (in good company with Jonatas…
Four (4 !) pyannoteAI speaker diarization models among the 10 most downloaded audio models on Hugging Face 🎉 (in good company with Jonatas…
Aimé par Théo Lepage
Expérience
Formation
-
Sorbonne Université
Doctor of Philosophy (Ph.D.) Artificial Intelligence
-
• Conducting research related to "Learning speech and speaker representations for robust speaker and language recognition"
• Supported by French ANR 'APATE' project (Forensic Deepfakes Detection Toolbox)
• Supervised by Dr. Réda Dehak and Pr. Thierry Géraud (LRE-EPITA) -
École pour l'informatique et les Techniques Avancées
Master of Engineering (M.Eng.) Computer Science
-
• International section
• Exchange semester in Spring 2019 at California State University Monterey Bay (CSUMB)
• Signal processing and machine learning (IMAGE major) + scientific research specialization (RDI major)
Licences et certifications
Publications
-
Additive Margin in Contrastive Self-Supervised Frameworks to Learn Discriminative Speaker Representations
Odyssey 2024
Self-Supervised Learning (SSL) frameworks became the standard for learning robust class representations by benefiting from large unlabeled datasets. For Speaker Verification (SV), most SSL systems rely on contrastive-based loss functions. We explore different ways to improve the performance of these techniques by revisiting the NT-Xent contrastive loss. Our main contribution is the definition of the NT-Xent-AM loss and the study of the importance of Additive Margin (AM) in SimCLR and MoCo SSL…
Self-Supervised Learning (SSL) frameworks became the standard for learning robust class representations by benefiting from large unlabeled datasets. For Speaker Verification (SV), most SSL systems rely on contrastive-based loss functions. We explore different ways to improve the performance of these techniques by revisiting the NT-Xent contrastive loss. Our main contribution is the definition of the NT-Xent-AM loss and the study of the importance of Additive Margin (AM) in SimCLR and MoCo SSL methods to further separate positive from negative pairs. Despite class collisions, we show that AM enhances the compactness of same-speaker embeddings and reduces the number of false negatives and false positives on SV. Additionally, we demonstrate the effectiveness of the symmetric contrastive loss, which provides more supervision for the SSL task. Implementing these two modifications to SimCLR improves performance and results in 7.85% EER on VoxCeleb1-O, outperforming other equivalent methods.
Other authorsSee publication -
Towards Supervised Performance on Speaker Verification with Self-Supervised Learning by Leveraging Large-Scale ASR Models
INTERSPEECH 2024
Recent advancements in Self-Supervised Learning (SSL) have shown promising results in Speaker Verification (SV). However, narrowing the performance gap with supervised systems remains an ongoing challenge. Several studies have observed that speech representations from large-scale ASR models contain valuable speaker information. This work explores the limitations of fine-tuning these models for SV using an SSL contrastive objective in an end-to-end approach. Then, we propose a framework to learn…
Recent advancements in Self-Supervised Learning (SSL) have shown promising results in Speaker Verification (SV). However, narrowing the performance gap with supervised systems remains an ongoing challenge. Several studies have observed that speech representations from large-scale ASR models contain valuable speaker information. This work explores the limitations of fine-tuning these models for SV using an SSL contrastive objective in an end-to-end approach. Then, we propose a framework to learn speaker representations in an SSL context by fine-tuning a pre-trained WavLM with a supervised loss using pseudo-labels. Initial pseudo-labels are derived from an SSL DINO-based model and are iteratively refined by clustering the model embeddings. Our method achieves 0.99% EER on VoxCeleb1-O, establishing the new state-of-the-art on self-supervised SV. As this performance is close to our supervised baseline of 0.94% EER, this contribution is a step towards supervised performance on SV with SSL.
Other authorsSee publication -
Experimenting with Additive Margins for Contrastive Self-Supervised Speaker Verification
INTERSPEECH 2023
Most state-of-the-art self-supervised speaker verification systems rely on a contrastive-based objective function to learn speaker representations from unlabeled speech data. We explore different ways to improve the performance of these methods by: (1) revisiting how positive and negative pairs are sampled through a "symmetric" formulation of the contrastive loss; (2) introducing margins similar to AM-Softmax and AAM-Softmax that have been widely adopted in the supervised setting. We…
Most state-of-the-art self-supervised speaker verification systems rely on a contrastive-based objective function to learn speaker representations from unlabeled speech data. We explore different ways to improve the performance of these methods by: (1) revisiting how positive and negative pairs are sampled through a "symmetric" formulation of the contrastive loss; (2) introducing margins similar to AM-Softmax and AAM-Softmax that have been widely adopted in the supervised setting. We demonstrate the effectiveness of the symmetric contrastive loss which provides more supervision for the self-supervised task. Moreover, we show that Additive Margin and Additive Angular Margin allow reducing the overall number of false negatives and false positives by improving speaker separability. Finally, by combining both techniques and training a larger model we achieve 7.50% EER and 0.5804 minDCF on the VoxCeleb1 test set, which outperforms other contrastive self supervised methods on speaker verification.
Other authorsSee publication -
Label-Efficient Self-Supervised Speaker Verification With Information Maximization and Contrastive Learning
INTERSPEECH 2022
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…
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.
Other authorsSee publication
Langues
-
Anglais
Capacité professionnelle générale
-
Français
Bilingue ou langue natale
Plus d’activités de Théo
-
Un de mes anciens étudiants SCIA vient d'obtenir son titre de Docteur, bravo Alae
Un de mes anciens étudiants SCIA vient d'obtenir son titre de Docteur, bravo Alae
Aimé par Théo Lepage
-
I'm excited to share that I have recently joined Resemble AI where I will focus on audio DeepFake detection.
I'm excited to share that I have recently joined Resemble AI where I will focus on audio DeepFake detection.
Aimé par Théo Lepage
-
La chute est spectaculaire : selon Les Echos, l'immobilier de bureaux en Ile de France vient de perdre un tiers de sa valeur en 2 ans, pour revenir à…
La chute est spectaculaire : selon Les Echos, l'immobilier de bureaux en Ile de France vient de perdre un tiers de sa valeur en 2 ans, pour revenir à…
Aimé par Théo Lepage
-
Now deployed at the Quinze-Vingts and Rothschild Foundation eye hospitals in Paris, real-time Doppler holography of the retina is set to become the…
Now deployed at the Quinze-Vingts and Rothschild Foundation eye hospitals in Paris, real-time Doppler holography of the retina is set to become the…
Aimé par Théo Lepage
-
We had the immense privilege of hosting Victor Miara from NVIDIA for our presentation on Large Language Models (LLM) with our team within the ESSEC…
We had the immense privilege of hosting Victor Miara from NVIDIA for our presentation on Large Language Models (LLM) with our team within the ESSEC…
Aimé par Théo Lepage
Autres profils similaires
Autres personnes nommées Théo Lepage (France)
-
Théo Lepage
En recherche d'un emploi flexible
-
Théo Lepage
Président chez SAS LE FUMOIR FRANCAIS
-
Theo Lepage
Responsable service comptabilité
-
Théo Lepage
Dessinateur / projeteur chez AIA Ingénierie
France : 6 autres personnes nommées Théo Lepage sont sur LinkedIn
Autres personnes nommées Théo Lepage