Voxtral TTS supplies the speaking component of Mistral AI's audio tools. The earlier Voxtral models accepted recordings for transcription and questions, as their release explains. TTS takes text and a reference voice to generate speech across nine supported languages. The March 2026 announcement describes using it alongside transcription and language models to produce spoken responses. It also reports transferring a reference voice's accent into another language, making voice adaptation part of the speech-generation step.
