Gabriel Synnaeve

Career

Gabriel Synnaeve's PhD work, supervised by Pierre Bessière, applied Bayesian modeling to StarCraft AI research. From 2013 to 2015 Synnaeve was a postdoctoral researcher at the École Normale Supérieure, modeling language acquisition and working on weakly supervised deep learning for acoustic modeling.³

Synnaeve joined Facebook AI Research as a postdoctoral researcher in 2015 and became a research scientist in 2016, working on end-to-end speech recognition and reinforcement learning benchmarks. From 2017 to 2019 Synnaeve was a research scientist and manager who tech-led a StarCraft: Brood War bot research project, and from 2019 to 2021 worked on robust and semi-supervised speech recognition and object detection with transformers, becoming a senior staff research scientist in 2021.³¹

Since 2021 Synnaeve has been a research scientist manager at FAIR, managing speech and audio work, audio compression and audio self-supervised learning while also doing individual research on code generation and translation. An October 2026 paper lists Synnaeve with Meta Superintelligence Labs.³²

Research and open-source work

With Ronan Collobert and Christian Puhrsch, Synnaeve co-authored Wav2Letter (2016), an end-to-end convolutional speech recognition system released in the wav2letter repository (since consolidated into Flashlight), and later co-authored the 2022 paper on Flashlight, a C++ machine learning library. Synnaeve was also a co-author of DETR (2020), which frames object detection as a direct set prediction problem with transformers.^59101112

In audio, Synnaeve co-authored EnCodec (2022), a neural audio codec, and MusicGen (2023), a text-conditioned music generation model, both released through Meta's AudioCraft library. In code generation, Synnaeve is the last author of Code Llama (2023) and a co-author of CWM (2025), an open-weights language model from the FAIR CodeGen team for research on code generation with world models.^6713414

Work and contributions

• Code Llama Open weights

  Last author of Code Llama: Open Foundation Models for Code (2023), a family of code models based on Llama 2 in 7B, 13B, 34B, and 70B sizes; the abstract says Code Llama is released under a permissive license that allows research and commercial use.⁴

• Wav2Letter and Flashlight Open-source code

  Co-author with Ronan Collobert and Christian Puhrsch of the Wav2Letter paper (2016) and a co-author of the Flashlight paper (2022). The wav2letter repository states that wav2letter++ is MIT-licensed, and the Flashlight repository is published under the MIT license.⁵⁹¹⁰¹¹

• EnCodec and MusicGen Open-source code

  Co-author of the EnCodec paper (2022) and a core-team author of the MusicGen paper (2023); both models are part of the AudioCraft repository, which releases its code under the MIT license and its model weights under CC-BY-NC 4.0.⁶⁷¹⁵¹³

Timeline

1. 2013Began a postdoctoral position at the École Normale Supérieure, working on language acquisition and acoustic modeling.³
2. 2015Joined Facebook AI Research as a postdoctoral researcher in speech recognition.³¹
3. Sep 11, 2016Co-authored Wav2Letter, an end-to-end ConvNet-based speech recognition system.⁵
4. 2017Became a research scientist and manager at FAIR, tech-leading a StarCraft - Brood War AI project.³
5. May 26, 2020Co-authored DETR, "End-to-End Object Detection with Transformers".¹²
6. 2021Became a research scientist manager at FAIR overseeing speech and audio work.³
7. Oct 24, 2022Co-authored EnCodec, "High Fidelity Neural Audio Compression".⁶
8. Jun 8, 2023Co-authored MusicGen, "Simple and Controllable Music Generation".⁷
9. Aug 24, 2023Last author of "Code Llama - Open Foundation Models for Code".⁴
10. Sep 30, 2025Co-authored CWM, an open-weights model for research on code generation with world models.¹⁴
11. Oct 6, 2026Co-authored "Agent Plasticity", listed with Meta Superintelligence Labs.²

Other documented roles

Roles the sources document that are not confirmed as current.

• Postdoctoral Researcher (2013–2015), École Normale Supérieure³¹

Where they work

• Meta 106 people in the wiki

Related people

• Michael AuliAlso at Meta
• Ronan CollobertAlso at Meta
• Joe SpisakAlso at Meta
• Alexei BaevskiAlso at Meta
• Ju-Chieh ChouAlso at Meta
• Dong YuSpeech and audio · ML frameworks and libraries
• Jonathan CohenLanguage models · Speech and audio
• Thomas WolfML frameworks and libraries · Language models
• Jeff DeanML frameworks and libraries · Language models
• Kai-Fu LeeSpeech and audio · Language models