French AI company Mistral has released a new open-source text-to-speech model that allows businesses to create voice agents for sales and customer engagement, putting Mistral in direct competition with players like ElevenLabs, Deepgram, and OpenAI.
French AI company Mistral has released a new open-source text-to-speech model that allows businesses to create voice agents for sales and customer engagement, putting Mistral in direct competition with players like ElevenLabs, Deepgram, and OpenAI.
The new model, called Voxtral TTS, supports 9 languages: English, French, German, Spanish, Dutch, Portuguese, Italian, Hindi, and Arabic.
“Our customers have long requested a speech generation model. That’s why we’ve developed a compact model that can run on smartwatches, smartphones, laptops, and other peripherals. It costs a fraction of anything on the market, yet delivers cutting-edge performance,” Pierre Stock, VP of Scientific Operations at Mistral AI, told TechCrunch.
Mistral said the new model can adapt its own voice with a sample of less than five seconds and capture characteristics such as subtle accents, intonations, logical stresses and individual speech tempos.
The model, built on the Ministral 3B, can easily switch between languages while retaining unique voice characteristics, which is extremely useful for dubbing or simultaneous interpretation. Stock emphasized that the company aimed to achieve a natural human sound, rather than a mechanical “robot voice.”
The model is designed to work in real time, according to the company. The Time-to-First-Audio (TTFA) indicator — the time from receiving input to the start of “speech” — is 90 ms for a 10-second sample of 500 characters. The model also has a real-time factor (RTF) of 6x, which means the ability to generate a 10-second audio clip in about 1.6 seconds.
