AI People Wiki

Builders & open source

Ronan Collobert

Research scientist, Machine Learning Research, Apple

Ronan Collobert is a research scientist in Apple's Machine Learning Research group and was previously a research scientist at Facebook AI Research, the Idiap Research Institute, and NEC Laboratories of America. Collobert's homepage lists a line of machine learning software: Torch, whose Torch7 release is distributed under a BSD license; SENNA, a natural language processing tagger; wav2letter, a toolbox for end-to-end speech recognition; and flashlight, a C++ machine learning framework that wav2letter is now part of. Collobert is first author of the Wav2Letter paper and last author of the Flashlight paper.1234

At a glance

Current role
Research scientist, Machine Learning Research
Company
Apple
Sector
Company
Work listed
2
Honors
1
Sources cited
7
Profile reviewed

Research

Collobert's research covers machine learning on structured data such as language, images and speech, with interests in semi-supervised, weakly supervised and structured learning and in large-scale learning. With Jason Weston, Collobert wrote "A Unified Architecture for Natural Language Processing" (ICML 2008), which trained one deep network on several language tasks at once and received ICML's Test of Time Award in 2018, and the follow-up "Natural Language Processing (Almost) from Scratch" (2011).15

Collobert later worked on computer vision and speech. "Recurrent Convolutional Neural Networks for Scene Labeling" (ICML 2014), with Pedro Pinheiro, applied recurrent convolutional networks to image labeling, and Wav2Letter (2016) proposed an end-to-end speech recognizer built from convolutional networks. Speech papers that followed include wav2letter++ (2019), a massively multilingual speech recognition model covering 50 languages (2020), and work on semi-supervised pseudo-labeling such as slimIPL (2021).13

Open-source work

Collobert's homepage describes a long line of machine learning software, beginning with SVMTorch, a support vector machine library written during doctoral work, and the Torch library through versions 3, 4, 5 and Torch7, which pairs the Lua scripting language with a C implementation under a BSD license. Collobert also released SENNA, a fast natural language tagger for part-of-speech tagging, chunking, named entity recognition and semantic role labeling, and co-wrote Torchnet (2016).1

More recent software includes wav2letter, a C++ toolbox for end-to-end speech recognition, and flashlight, a modern C++ machine learning framework into which wav2letter has been folded. Collobert is last author of the 2022 Flashlight paper; flashlight is published under the MIT license, and the wav2letter README describes wav2letter++ as MIT-licensed.1467

Work and contributions

  • flashlight and wav2letter Open-source code

    Collobert's homepage presents flashlight, a C++ machine learning framework, and wav2letter, an end-to-end speech recognition toolbox now part of flashlight. Collobert is first author of the Wav2Letter paper (2016) and last author of the Flashlight paper (2022); flashlight is published under the MIT license, and the wav2letter repository states that wav2letter++ is MIT-licensed.13467

  • Torch Open-source code

    Collobert's homepage lists Torch, a machine learning library whose last version, Torch7, pairs the Lua scripting language with a C implementation and is distributed under a BSD license, along with the earlier Torch 3, 4, and 5.1

Timeline

  1. Presented "A Unified Architecture for Natural Language Processing" with Jason Weston at ICML 2008.15
  2. Published "Natural Language Processing (Almost) from Scratch" and the Torch7 paper.1
  3. Released version 3.0 of the SENNA natural language processing tagger.1
  4. Presented "Recurrent Convolutional Neural Networks for Scene Labeling" at ICML 2014.1
  5. Posted "Wav2Letter", an end-to-end ConvNet-based speech recognition system, as first author.3
  6. Co-authored "wav2letter++", describing the open-source speech recognition system, at ICASSP.1
  7. Co-authored "Massively multilingual ASR", a single speech recognition model for 50 languages.1
  8. Posted the Flashlight machine learning framework paper as last author.4

Other documented roles

Roles the sources document that are not confirmed as current.

Honors and recognition

  • ICML Test of Time Award ("A Unified Architecture for Natural Language Processing") · International Conference on Machine Learning, 20185

Where they work

  • Apple 27 people in the wiki

Suggest a correction or removal

Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.

References

7 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026

  1. Ronan Collobert — homepageRonan Collobert · Official page · accessed Oct 21, 2026
  2. andresy (Ronan Collobert) on GitHub (company Apple)GitHub · Official page · accessed Oct 21, 2026
  3. Wav2Letter: an End-to-End ConvNet-based Speech Recognition System (arXiv 1609.03193)arXiv · Paper · published Sep 11, 2016 · accessed Oct 21, 2026
  4. Flashlight: Enabling Innovation in Tools for Machine Learning (arXiv 2201.12465)arXiv · Paper · published Jan 29, 2022 · accessed Oct 21, 2026
  5. ICML 2018 AwardsInternational Conference on Machine Learning · Official page · accessed Oct 21, 2026
  6. flashlight/flashlight repository (MIT license)Flashlight (GitHub repository) · Repository · accessed Oct 21, 2026
  7. flashlight/wav2letter README and licenseFlashlight (GitHub repository) · Repository · accessed Oct 21, 2026