Updated
Updated · KDnuggets · May 1
Mistral AI releases Voxtral TTS open-weight speech model
Updated
Updated · KDnuggets · May 1

Mistral AI releases Voxtral TTS open-weight speech model

11 articles · Updated · KDnuggets · May 1
  • The 4-billion-parameter system supports nine languages, clones voices from three seconds of audio and runs on local hardware, with commercial access also available through Mistral's API.
  • Mistral says Voxtral TTS delivers 70ms model latency, about 100ms time-to-first-audio and 9.7x real-time generation, targeting conversational AI, customer support, translation and accessibility tools.
  • The model's weights are available under a CC BY-NC 4.0 non-commercial licence, while paid API use costs $0.016 per 1,000 characters and commercial self-hosting requires separate licensing.
Mistral is betting its future on an open model. Will this strategy truly disrupt the established AI voice market leaders?
Voxtral TTS offers a self-hosted alternative to pricey APIs. But what are the hidden operational costs for businesses?
Mistral's AI clones voices in seconds. Can new US and EU laws actually prevent a crisis of deepfake fraud?