Seamless Communication

Seamless Communication is Meta's family of AI models for developers and researchers building multilingual speech applications. It covers translation and transcription, with dedicated models for conversations and expressive speech.

• Speech and text translation: SeamlessM4T translates spoken or written input into speech or text and also transcribes speech.
• Translation while speaking: SeamlessStreaming produces translated speech or text from incoming audio before the speaker finishes, and supports streaming transcription.
• Voice style preservation: SeamlessExpressive carries aspects of a speaker's voice style, pace and pauses into translated speech.
• Expressive streaming: The combined Seamless model brings voice style preservation to streaming speech-to-speech translation.

The models run through Meta's fairseq2 library, whose prebuilt packages support Linux x86-64 and Apple Silicon Macs. SeamlessM4T also works with Hugging Face Transformers. A smaller SeamlessM4T model targets on-device inference.