French AI startup Mistral AI has unveiled Voxtral, a new open-source model designed for real-time speech-to-speech translation. The model is notable for its speed, reportedly processing audio 2-3 times faster than leading competitors like OpenAI’s Whisper, while maintaining high accuracy across several languages. Voxtral operates by directly translating speech to speech without an intermediate text step, which contributes to its efficiency. The release is seen as a significant challenge to the dominance of larger, well-funded AI labs, demonstrating that smaller, specialized companies can innovate in high-performance AI applications. The model is available for researchers and developers to test and build upon. Read the full article at: https://www.wired.com/story/mistral-voxtral-real-time-ai-translation/