Changhan Wang

Meta Superintelligence Labs

[Changhan Wang @ChanghanWang]

@ParcolletT @PatrickPlaten @MetaAI No worries! Let me know if you have further questions

[Changhan Wang @ChanghanWang]

@ParcolletT @PatrickPlaten @MetaAI Our models are in line 3 and 4. We may release w2v2 transducer models in the future.

[Changhan Wang @ChanghanWang]

@ParcolletT @PatrickPlaten @MetaAI No, they were trained with CTC loss.

[Changhan Wang @ChanghanWang]

@PatrickPlaten @MetaAI credit to @sravyapopuri388 who implemented and trained these awesome models

[Juan Pino @juanmiguelpino]

We are currently hiring for 2 RS/RE positions as well as 1 Postdoctoral position. We are conducting research on exciting topics related to speech-to-speech translation with the aim to solve live multilingual communication. Please reach out if you are interested to learn more!

[AK @_akhaliq]

.@Gradio Demo for fairseq S2S: speech-to-speech translation models now on @huggingface Spaces
github: https://t.co/Q1r7oxlclv
Huggingface Space: https://t.co/RH5DxAen8N

[Juan Pino @juanmiguelpino]

We're happy to announce the 3rd edition of the simultaneous shared task at @iwslt ! Novelty this year: en-de/en-ja/en-zh, en-de manual eval and human interpretation benchmark, text-to-text and speech-to-text evaluated together. Consider participating! https://t.co/RsN0ogUdas

[AI at Meta @AIatMeta]

We've taken another step toward eliminating language barriers. In collaboration with @huggingface, we recently released 4 direct speech-to-text models that translate English to Arabic, Catalan, German & Turkish. Try out the models for yourself: https://t.co/mKSGLwYnBn

[Changhan Wang @ChanghanWang]

Check out our IWSLT winning recipe https://t.co/BnTNKwMkBE and our #ACL2021NLP paper https://t.co/075xQrTPdu !

[AI at Meta @AIatMeta]

We’re proud to announce that we won 1st place at this year’s annual multilingual speech translation competition hosted by @iwslt. Read the details on building a state-of-the-art speech translation system: https://t.co/AWvMwvKUoO

[Changhan Wang @ChanghanWang]

Happy to share our large-scale multilingual speech corpus, VoxPopuli:
- 400K-hour unlabelled speech in 23 languages
- 1.8K-hour transcribed speech in 16 languages
- 17.3K-hour speech-to-speech interpretation data in 15x15 directions

https://t.co/P1vH4jARCR

#ACL2021NLP

[AI at Meta @AIatMeta]

To help #NLP researchers create advanced translation & speech recognition systems for more languages spoken around the world, Facebook AI is releasing VoxPopuli. This data set provides 400K hours of unlabeled speech data in 23 languages. Learn more: https://t.co/yipgc4VFdb

[laurent besacier @laurent_besacie]

@formiel @aclmeeting @juanmiguelpino @ChanghanWang @thoma_gu @didier_schwab Our #ACL2021 work on adapters for speech translation will be presented today 9am UTC (11am CET) during poster session 3D: Speech and Multimodality. See you on gathertown for the poster ! Paper: https://t.co/6KPpxO44Im

[Xian Li @xl_nlp]

Multilingual Speech Translation from Efficient Finetuning of Pretrained Models. A simple recipe for SoTA speech translation for 36 pairs. https://t.co/H23nDIcApA session oral 3D with @ChanghanWang Yun Tang @juanmiguelpino @MichaelAuli and other colleagues at FAIR #ACL2021NLP 2/n

[Changhan Wang @ChanghanWang]

How much can we improve speech translation without adding expensive labeled data? We achieve SOTA performance on CoVoST 2 (average +2.6 BLEU) by combining several techniques to leverage large-scale unlabeled speech/text data.

https://t.co/5wrmmLYHsH
Fairseq models coming soon

[Alexis Conneau @alex_conneau]

Happy to share our work "Large-Scale Self- and Semi-Supervised Learning for Speech Translation"

New state of the art (+2.6 BLEU on average) on CoVoST-V2 with a simple approach, lot of unannotated data and less supervision

Paper: https://t.co/SqmTNz0QbV
Models coming soon

[IWSLT @iwslt]

We're happy to announce the start of the four challenge tracks for the 2021 evaluation campaign!

🗓Feb 1: Training data released
🗓Apr 5-16: Evaluation period

See more at https://t.co/8ynQZdOwqY and register your interest below: https://t.co/To4oPbTRhG

[AI at Meta @AIatMeta]

We've just released our award-winning news translation models for #WMT2020. This features low-resource languages pairs: Tamil<->English and Inuktitut<->English Download the model here:
https://t.co/LrZYC8DGHa

[AI at Meta @AIatMeta]

We're sharing our work on improving translation for low-resource languages (Tamil and Inuktitut) at #wmt2020. By combining various techniques, we improved translation quality for these underserved languages. Tune in tomorrow at 12pm EST. Here's the paper: https://t.co/POnAmjY73z

[Juan Pino @juanmiguelpino]

Check out @xutai_ma's #EMNLP2020 demo today 6pm Pacific! SimulEval is a new toolkit to standardize latency measurement for simultaneous speech and text translation. Paper: https://t.co/5Sx9VrWNac Code: https://t.co/3ZvkbTUfYR with Javad Dousti, @ChanghanWang, @thoma_gu

[Hang Le @formiel]

Happy to share our recent work on a new architecture called the dual-decoder Transformer for joint speech recognition and multilingual speech translation (oral presentation @coling2020).

Paper: https://t.co/uamKw0h2Ht
Code: https://t.co/XSuDBz62dF

[Juan Pino @juanmiguelpino]

Check out https://t.co/T2hQPk1qH6: summary of the simultaneous task at 1:39:36, awesome panel on offline speech translation at 2:43:46 and another awesome panel on simultaneous speech translation at 3:45:55

[Juan Pino @juanmiguelpino]

We're looking for a postdoctoral researcher to work on speech to speech translation at @facebookai Please apply at https://t.co/J2mqXxzmtG and/or DM me