Voxtral releases open-weight multilingual TTS model

Mistral AI has launched Voxtral TTS (Voxtral-4B-TTS-2603), a lightweight, open-weight text-to-speech (TTS) model. This new model is designed for flexible deployment and broad adoption, emphasizing voice cloning, open weights, and natural, accented speech delivery.

Voxtral supports expressive speech generation across nine languages and can handle up to 30 minutes of audio at a time for tasks like transcription and long Q&A chats. It is positioned as an advancement in AI voice technology, aiming for widespread adoption due to its flexibility.