French developer Kyutai's Moshi AI voice assistant, with lifelike conversations, outperformed OpenAI in Voice Mode.

From Future plc: 2024-07-05 18:00:31

OpenAI’s delay of ChatGPT’s Voice Mode disappointed fans, but French developer Kyutai introduced Moshi, a real-time voice AI assistant. Moshi offers lifelike conversations, powered by large language models like ChatGPT. It can handle multiple accents, emotions, and speaking styles concurrently.

Kyutai fine-tuned Moshi with over 100,000 synthetic dialogues using Text-to-Speech technology to mimic human communication nuances. The AI integrates text and audio training, optimized for offline use to prioritize privacy and security. A demo of Moshi is available online, showcasing its capabilities.

Moshi will be an open-source project, facilitating further innovation and addressing concerns about AI safety and ethics. Kyutai is developing AI audio identification features for Moshi to enhance accountability and traceability. As Moshi evolves, it could influence other voice-enabled AI assistants or accelerate the integration of large language models into existing platforms.



Read more at Future plc: This new AI voice assistant beat OpenAI to one of ChatGPT’s most anticipated features