Mistral AI Launches Voxtral TTS for Enterprise Voice Generation

Mistral AI has launched a new open source text-to-speech model, marking the French company’s latest move to provide a comprehensive suite of voice products for enterprise use. The model, called Voxtral TTS, is designed for applications such as voice AI assistants, customer support, and sales engagement. By entering the speech generation market, Mistral now competes directly with established players including ElevenLabs, Deepgram, and OpenAI.

Performance and Technical Capabilities

Based on the Ministral 3B architecture, Voxtral TTS is engineered for efficiency and real-time performance. Pierre Stock, VP of science operations at Mistral AI, noted that the model is small enough to run on edge devices, including laptops, smartphones, and smartwatches. According to the company, the model achieves a time-to-first-audio (TTFA) of 90 ms for a 10-second sample of 500 characters. Furthermore, it boasts a real-time factor (RTF) of 6x, allowing it to render a 10-second audio clip in approximately 1.6 seconds.
The model supports nine languages: English, French, German, Spanish, Dutch, Portuguese, Italian, Hindi, and Arabic. Mistral designed the technology to maintain voice characteristics, such as subtle accents, intonations, and inflections, even when switching between languages. This capability is intended to support complex use cases like real-time translation and dubbing while avoiding a robotic sound.

Customization and Enterprise Strategy

A key feature of Voxtral TTS is its ability to adapt a custom voice using a sample of less than five seconds. Mistral is positioning the model’s open source nature and customization options as a primary advantage for enterprises, allowing businesses to tune the voice output to meet specific requirements.
This release follows the launch of two transcription models earlier in the year, which were designed for both large batch processing and low-latency, real-time use cases. Looking ahead, Mistral aims to build an end-to-end platform capable of handling multimodal streams of input and output, including text, audio, and images. According to Stock, the company believes that an end-to-end agentic system supporting audio will provide significantly more information and utility for enterprise customers.