Kyutai Research Labs today released the Moshi AI voice assistant in Paris, an AI voice assistant that can have natural conversations with humans, comparable to the voice capabilities of OpenAI’s GPT 4o. The voice assistant was developed by an 8-person team over 6 months and has unique emotional and AI interaction capabilities.
Kyutai will make Moshi’s code and model weights publicly available, enabling researchers and developers to freely use, improve, and extend the technology.
Features of Moshi
Voice interaction capabilities
Natural Conversation: Moshi is capable of natural, fluent and expressive voice conversations, simulating the way humans communicate with each other.
Emotional expression: Its text-to-speech (TTS) capability is excellent in emotional expression and can present rich emotional changes.
Versatile Applications
Coach and Companion: Moshi can be used as a personal coach or companion, providing guidance, support and interaction to help users get personalized advice and companionship in different situations.
Role-playing: Possessing the ability to play roles, being able to demonstrate strong creativity and flexibility during the interaction process, which is suitable for scenarios such as games and education.
Real-time interaction
Instant Response: During demonstrations and interactions, Moshi can quickly respond to users’ voice commands and questions, providing a smooth interactive experience.
Efficient multimodal processing
Multimodal learning and reasoning: Moshi has the ability to process and understand multiple types of content (such as text, sound, images, etc.), and can effectively learn and reason between different content.
Technology openness
Code and Model Publicity: Kyutai will make Moshi’s code and model weights publicly available, enabling researchers and developers to freely use, improve, and extend this technology.
Local operation: Moshi can be installed and run locally to ensure security and stability in an offline environment.
Kyutai is a non-profit laboratory dedicated to open research in AI, founded in November 2023 by the Iliad Group, CMA CGM and Schmidt Sciences. The startup team consists of six top scientists, all of whom have worked in large American technology labs. Kyutai continues to recruit top talent and also offers internships to research master’s students. The team now has 12 members and will start the first doctoral dissertation research at the end of the year. Research explores new general-purpose models with high capabilities. The lab is currently working specifically on multimodal models, i.e. models that are able to learn and reason with different types of content (text, sound, images, etc.). All developed models, software and the technological know-how that enabled their creation will be shared free of charge. To carry out its work and train its models, Kyutai relies in particular on Nabu 23 supercomputing nodes provided by Scaleway, a subsidiary of the Iliad Group.
