Mathieu Fontaine (@MathieuFontai19) profile picture

Assiociate Professor at Telecom Paris. Working on machine listening and mixed reality

Palaiseau, France
Joined May 2019
Photo by Mathieu Fontaine (@MathieuFontai19)Photo by Mathieu Fontaine (@MathieuFontai19)Photo by Mathieu Fontaine (@MathieuFontai19)Photo by Mathieu Fontaine (@MathieuFontai19)Photo by Mathieu Fontaine (@MathieuFontai19)Photo by Mathieu Fontaine (@MathieuFontai19)Photo by Mathieu Fontaine (@MathieuFontai19)Photo by Mathieu Fontaine (@MathieuFontai19)Photo by Mathieu Fontaine (@MathieuFontai19)Photo by Mathieu Fontaine (@MathieuFontai19)
Mathieu Fontaine retweeted
Every language in Monsoon V1, in one spin of the globe. 50 languages. 23 countries. Spontaneous speech, not read prompts. If your ASR works in English and falls apart everywhere else, this is the data it was missing.
3
45
15
91
173,460
Language modeling is ambiguous. If you ask someone to caption an image, describe an audio clip, or to translate a sentence, you’ll get several valid outputs. In this work, we handle this with a winner-takes-all training objective that encourages diverse, plausible predictions.
Replying to @valeoai
1/ Multiple Choice Learning of Low-Rank Adapters for Language Modeling TL;DR: LoRA heads + Winner-Takes-All loss → diverse, plausible LLM outputs 📄 arxiv.org/abs/2507.10419 💻 github.com/Victorletzelter/L… @VLetzelter, H. Malard, M. Fontaine, G. Richard, S. Essid, @abursuc, @ptrkprz
1
3
1
9
1,444
This is the final work from my PhD, in collaboration with @abursuc @Mldhug @MathieuFontai19 @.grichard @.slimessid @ptrkprz at @valeoai and @telecomparis Our poster presentation is Thu, Jul 9, 2026 • 10:30 AM – 12:15 PM KST.
1
2
4
134
Interested in ill-posed learning tasks, uncertainty prediction, conditional density estimation or multi-head deep neural networks ? In our new paper, accepted at #ICML24, we tackle these challenges by exploring the Winner-Takes-All (WTA) training scheme. [1/n]
1
10
2
35
5,592
Mathieu Fontaine retweeted
Our paper entitled "Neural Blind Source Separation and Diarization for Distant Speech Recognition" is accepted to Interspeech 2024! Our neural FCASA is a method to jointly separate and diarize speech mixtures without supervision by isolated signals. arxiv.org/abs/2406.08396
2
16
3
59
24,703
Replying to @AdaSp
@adasp group from ⁦⁦@telecomparis⁩ @ieeeICASSP⁩ 2014, Seoul, Korea
2
2
13
776
Mathieu Fontaine retweeted
Resilient Multiple Choice Learning: A learned scoring scheme with application to audio scene analysis by @VLetzelter @MathieuFontai19 @Mickael_Chen P. Pérez, S. Essid, G. Richard tl;dr: extend MCL for cond. distrib. estimation in regression #neurips2023 arxiv.org/abs/2311.01052
1
1
4
297
Mathieu Fontaine retweeted
Today, we announce Ego-Exo4D, the largest and most diverse multi-view dataset, showing human experts around the world performing a core set of skilled activities, w/ unprecedented multi-modality, novel new video-language resources, and rich annotations
7
97
55
506
304,801
We released torchiva, a python package for multi-channel source separation with independent vector analysis (IVA). It allows to mix traditional signal processing with DNN source model! 🥳 Give it a go with the handy separate command! Code by me and @kohei_1979 🧵1/2
1
25
108
I released fast_bss_eval, a python package to compute SDR/SIR/SAR metrics for the evaluation of blind source separation algorithms! 1/4 pip install fast-bss-eval - numpy/torch ok #⃣🔥 - fast! 🚄 - GPU+differentiable 🎉 doc: fast-bss-eval.readthedocs.io… repo: github.com/fakufaku/fast_bss…
3
20
1
72
📢 DCASE Challenge 2021 results have been published! 🎉🥳🎉 More info on each task webpage 👇 dcase.community/challenge202… Thanks again to all the participants and task organizers! 👏
Mathieu Fontaine retweeted
A friend of a friend sent me this...
7
35
7
145
Now you can officially register to participate in the #MDXChallenge 2021! Just follow the link and click the "Participate" button! The first round of the competition will start in two days! aicrowd.com/challenges/music…
1
5
16
Mathieu Fontaine retweeted
arXiv is free to use, but not free to operate. If everyone who visits arXiv this month gives just one dollar, we would meet our annual operating budget and be able to save for future financial sustainability. Considering donating here: https://teapawt.net/t.co/TbYAssKDsh
14
390
60
971
🎉 DESED now available on pypi ! 🎉 Makes it easy to download: ✔️ DESED dataset ✔️ FUSS ✔️ FSD50K Use all the above to generate synthetic mixtures: go check the 2021 example! 🔥🔥 github.com/turpaultn/DESED
3
14
Mathieu Fontaine retweeted
Slides and notebooks for my new course on #Bayesian methods for #MachineLearning at @centralesupelec 👇 github.com/sleglaive/Bayesia…
8
18
Flow-Based Independent Vector Analysis for Blind Source Separation - IEEE Journals & Magazine ieeexplore.ieee.org/document…
2
10
23