1. Meta has launched the AI translation model SeamlessM4T, which supports speech-to-text, speech-to-speech, text-to-text, and text-to-speech for nearly 100 languages. 2. SeamlessM4T's speech recognition capabilities have reached human-level performance, significantly improving its ability to handle background noise and varied speech. 3. Meta has not only open-sourced the SeamlessM4T model but also made the key training dataset SeamlessAlign publicly available.

1. Meta's newly open-sourced translation model SeamlessM4T supports translation in 100 languages and voices. 2. SeamlessM4T integrates the advantages of multiple translation models, including NLLB and MMS. 3. SeamlessM4T is trained on 270,000 hours of aligned speech and text data.