Skip to content
AI News HubLIVE
Summary only

Closing the ‘Expressivity Gap’: How Mistral’s Voxtral TTS is Redefining Multilingual Voice Cloning with a Hybrid Autoregressive and Flow-Matching Architecture

Mistral AI releases Voxtral TTS, a ~4B parameter multilingual text-to-speech model that combines autoregressive generation and flow-matching to close the expressivity gap. It achieves a 68.4% win rate over ElevenLabs Flash v2.5 in zero-shot voice cloning across 9 languages, generates natural speech from as little as 3 seconds of reference audio, and serves over 30 concurrent users on a single H200 at sub-600ms latency. The article details the architecture, post-training with DPO, and use cases.

SourceMarkTechPostAuthor: Asif Razzaq
Closing the ‘Expressivity Gap’: How Mistral’s Voxtral TTS is Redefining Multilingual Voice Cloning with a Hybrid Autoregressive and Flow-Matching Architecture
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions

Summary only

Under the source content permission, this page provides an editorial summary, key points, and a link to the original article.

Key points and analysis

Article intelligence

EngineersAdvanced

Key points

  • Voxtral TTS uses a hybrid architecture: an autoregressive decoder for semantic coherence and a flow-matching transformer for acoustic expressivity. It outperforms competitors particularly in low-resource languages like Hindi and Spanish.
  • The model supports zero-shot cross-lingual voice adaptation without fine-tuning and is available as open weights (CC BY-NC 4.0) or API at $0.016 per 1,000 characters.
  • Post-training with DPO reduces word error rates and improves naturalness, though Hindi WER regressed slightly.

Highlights and analysis are generated automatically and may contain errors. Check the original source.

More recent stories on this topic · Models