Mistral launches Voxtral TTS, completing its open-source speech pipeline
Mistral launched a text-to-speech model that clones a voice from 3 seconds of audio and runs on 3GB of RAM.
Mistral launched Voxtral TTS today, a new open-weight text-to-speech model that rounds out the company’s full speech pipeline.
Voxtral started in July 2025 as an audio understanding family — models that could transcribe, translate, and answer questions from voice input. Early 2026 brought specialized transcription variants. Voxtral TTS is the output side of that equation.
The model is around 3 to 4 billion parameters and runs on approximately 3GB of RAM, making it viable on edge devices.
- Voice cloning: Adapts to a new voice from just 3 to 5 seconds of reference audio, cross-lingual while preserving accent and tone.
- Latency: Around 70 to 90ms time-to-first-audio.
- Languages: 9 languages and dialects with emotional prosody support.
- Performance: Mistral claims stronger human preference scores than ElevenLabs Flash v2.5.
The weights are available on Hugging Face under CC BY-NC 4.0.
The full Voxtral stack now covers transcription, real-time speech understanding, and synthesis in one open-weight family.
For developers building voice agents or speech applications who want to avoid stitching together Whisper, a separate LLM, and a third-party TTS service, that is a meaningful simplification.
Bottom line: Mistral quietly built one of the most complete open-source speech stacks available.
Source: Mistral AI / Hugging Face
If you need on-demand GPUs for training, fine-tuning, inference, or running open-source models, give RunPod a try.
- Available hardware: H100, H200, A100, L40S, RTX 4090, RTX 5090, and 30+ more
- Cost: significantly cheaper than AWS or GCP, billed per second, no contracts
- Setup: spins up in under a minute, 30+ regions worldwide
Get the core business tech news delivered straight to your inbox. We track AI, automation, SaaS, and cybersecurity so you don't have to.
Just read what you want, and be done with it.
