AI language translation models by Meta

I'm reposting from here[0][1][2] because there was no discussion

The original title is "Introducing a suite of AI language translation models that preserve expression and improve streaming". I've shortened it here so that it's more obvious that these models are from Meta (and worth discussing). The code is on GitHub:

https://github.com/facebookresearch/seamless_communication

From the GitHub: SeamlessM4T is our foundational all-in-one Massively Multilingual and Multimodal Machine Translation model delivering high-quality translation for speech and text in nearly 100 languages. SeamlessM4T models support the tasks of:

    - Speech-to-speech translation (S2ST)
    - Speech-to-text translation (S2TT)
    - Text-to-speech translation (T2ST)
    - Text-to-text translation (T2TT)
    - Automatic speech recognition (ASR)

A wild guess: Meta going to be working on embedding this into hardware as fast as possible so that you can chat to anyone on the planet, no matter which language they speak

Related papers:

- Zero-shot Expressive Speech-to-Speech Translation: https://ai.meta.com/research/publications/sonar-expressive-z...

- Efficient Monotonic Multihead Attention: https://ai.meta.com/research/publications/efficient-monotoni...

- Multilingual Expressive and Streaming Speech Translation

[0] https://news.ycombinator.com/item?id=38477728

[1] https://news.ycombinator.com/item?id=38479724

[2] https://news.ycombinator.com/item?id=37232489