French AI company Mistral has released an open-source text-to-speech (TTS) model called Voxtral TTS, TechCrunch reported on March 26.
The model can be used for voice AI assistants or customer support solutions and will compete directly with ElevenLabs, Deepgram and OpenAI.
Voxtral TTS supports 9 languages, including English, French, German, Spanish, Dutch, Portuguese, Italian, Hindi and Arabic. Pierre Stock, vice president at Mistral AI, said the company developed a small voice model that runs on edge devices such as smartwatches, smartphones and laptops at customer request. He said it offers state-of-the-art performance while lowering costs compared with existing models.
Voxtral TTS can generate custom voices from voice samples of under 5 seconds and can capture subtle differences in intonation, stress and pronunciation. It can be used for dubbing or real-time translation by preserving voice characteristics when switching between languages. Mistral said it developed the model with the goal of producing a human-like voice rather than a robotic one.
Chi-gyu Hwang (황치규)
delight@d-today.co.kr