Voxtral TTS

Voxtral TTS is Mistral AI's open-source text-to-speech model, released March 26, 2026, as the speech-synthesis counterpart to the Voxtral audio-understanding line. It covers nine languages, English, French, German, Spanish, Dutch, Portuguese, Italian, Hindi, and Arabic, with zero-shot voice cloning from a 3-to-30-second reference clip, emotion steering, and natural prosody aimed at conversational voice agents.

Mistral's emphasis is compactness: the model is designed to run not only in the cloud but on smartphones, laptops, and other edge devices, a differentiator against cloud-only commercial TTS. It arrived alongside Cohere Transcribe on the same day, marking a multi-vendor push into open audio AI in early 2026.