@MichaelAuli

Research Scientist @FacebookAI

 San Francisco, CA
michaelauli.github.io/
Joined October 2020
  • Tweets87
  • Following154
  • Followers2.8K
23Media
Great progress by the team at Ello! Child speech is a very challenging but rewarding problem to work on! Looking forward to see the impact of Ello 2.0 and beyond!
Catalin Voss
@CatalinVossJul 7
Everyone's asking what AI can do for their job. We asked if it could teach a 6-year-old. Introducing Ello 2.0: reading, math & more for ages 4-9. Free tier for all. If we want AI to force-multiply humanity, let's start with teaching our youngest.
1
13
1,750
Michael Auli retweeted
David AI
@withdavidai8 Oct 2025
David AI has raised a $50M Series B from Meritech and NVIDIA to establish the data layer for audio AI. Audio is the front-end interface for real-world AI. At David AI, we’re creating the data that powers the models bringing these use cases to life. withdavid.ai/news/announcing…

Announcing Our $50M Series B Led by Meritech | David AI Blog

withdavid.ai
21
20
12
147
171,773
We are releasing MMS Zero-shot: a model to transcribe the speech of almost any language using only a small amount of unlabeled text in the new language. Paper: arxiv.org/abs/2407.17852 Demo: huggingface.co/spaces/mms-me… Code/model: github.com/facebookresearch/…
6
32
6
197
95,607
Michael Auli retweeted
kyutai@kyutai_labs17 Nov 2023
Our founding team is covering many AI fields from vision, with Patrick Pérez and Hervé Jégou (@hjegou) to LLMs with Edouard Grave (@EXGRV), audio with Neil Zeghidour (@neilzegh) and Alexandre Défossez (@honualx) and infra with Laurent Mazaré (@lmazare).
7
34
19
217
95,629
Here is an exciting new brain computer interface challenge taking ASR to the next level! Consider participating!
Frank Willett@WillettNeuro23 Aug 2023
Replying to @WillettNeuro
We are publicly releasing all data and code, and are hosting a machine learning competition! Can you do better than us at translating neural activity into text? 2/3 eval.ai/web/challenges/chall…

EvalAI: Evaluating state of the art in AI

EvalAI is an open-source web platform for organizing and participating in challenges to push the state of the art on AI tasks.

eval.ai
7
1,540
The MMS TTS systems are now on HuggingFace! This is TTS for 1,107 languages.
Sanchit Gandhi@sanchitgandhi991 Sep 2023
🤗 Transformers just got 1100+ new TTS checkpoints 🚀 You can now run any of @MetaAI's MMS TTS checkpoints using the Transformers library in 3 lines of code⚡️ MMS is a the largest democratizer of TTS globally to date 🌎 Try it in your language now: colab.research.google.com/gi…
3
12
1,950
Excited to be at ICML 2023 where we will present our work on data2vec 2.0 on Wednesday (talk) and Thursday (poster): icml.cc/virtual/2023/oral/25… @alexei_baevski @arunbabu1234 @mhnt1580
1
5
37
5,362
The MMS models are now available on HuggingFace!
Vaibhav (VB) Srivastav
@reach_vb21 Jun 2023
Meta AI's recently released "Massively Multilingual Speech" (MMS) model is a huge step forward towards democratizing Speech to every corner of the globe. In addition, it might also play a significant role in preserving global linguistic diversity. huggingface.co/papers/2305.1… 👇

Paper page - Scaling Speech Technology to 1,000+ Languages

Join the discussion on this paper page

huggingface.co
1
13
1,234
New work! The Massively Multilingual Speech (MMS) project scales speech technology to 1,100-4,000 languages using self-supervised learning with wav2vec 2.0. Paper: research.facebook.com/public… Blog: ai.facebook.com/blog/multili… Code/models: github.com/facebookresearch/…
14
120
24
444
182,315
more replies
Compared to OpenAI Whisper, the multilingual ASR model supports 11x more languages but has less than half the average error rate on 54 languages of FLEURS. The model is also trained on a fraction of the labeled data (45K vs. 680K hours).
1
1
2
22
2,710
We also find that scaling multilingual ASR to this many languages only results in a very small performance degradation of 0.4 character error rate while increasing the number of supported languages by 18x (61 -> 1,107)
2
11
1,836
Michael Auli retweeted
Jiatao Gu
@thoma_gu26 Dec 2022
We (@melbayad @MichaelAuli @EXGRV) have also worked on a very similar approach to adaptively change the depth of decoding for MT date back to ICLR2020 openreview.net/forum?id=SJg7…
Jay Hack
@mathemagic1an26 Dec 2022
Current LLMs expend the same amount of computation on each token they generate. But some predictions are much harder than others! With CALM, the authors redirect computational resources to "hard" inferences for better perf (~50% speedup) ai.googleblog.com/2022/12/ac… Here's how 👇
2
20
4,749
New work on efficient self-supervised learning: data2vec 2.0 pre-trains vision models 16.4x faster than the most popular existing algorithm. Blog: ai.facebook.com/blog/ai-self… Paper: ai.facebook.com/research/pub… Code/models: github.com/facebookresearch/… with @ZloiAlexei @mhnt1580 @arunbabu1234
4
61
11
303
more replies
data2vec 2.0 pre-trains vision models to the same accuracy as MAE in 3 hours on a distributed training setup, or under 100 hours on a single A100 (not accounting for network overhead). For speech it trains 10.6x as fast as wav2vec 2.0 and for NLP 2x as fast as RoBERTa.
1
2
10
The key ingredients are the use of contextualized representations which enable a fast self-supervised learner. We improve efficiency by masking each training example multiple times which amortizes the effort to compute the teacher network. We also use MAE-style sample encoding.
2
4
Join us for the Computer Perception virtual workshop at ECCV this Sunday (computerperception.github.io…). We will have keynotes from a wonderful set of speakers @orussakovsky, @AudeOliva, Matt Botvinick, @MichaelAuli, @dyamins, @JitendraMalikCV + talk about how to test perception.
5
9
2
31
I will be at Interspeech next week! Looking forward to attending my first in-person conference for a long time and talking to people about our papers (XLS-R, unsupervised TTS, zero-shot phone recognition, more efficient wav2vec, etc.)
1
21
Load more