AI Chatbot Moshi: A new artificial intelligence chatbot called Moshi can recognize the tone of your speech, answer more quickly than ChatGPT’s planned Advanced speech Mode, and be interrupted.
A new AI-powered chatbot named “Moshi” from French startup Kyutai contains characteristics akin to ChatGPT’s now-delayed “Advanced Voice Mode” GPT-4o. Moshi is able to decipher and understand your speech tone. It’s offline use is also possible.
Introducing Moshi: An Advanced AI Chatbot with Speaking Styles and 70 Emotions, Powered by Helium LLM
The chatbot, which is presently accessible to everyone, is built on a 7B parameter large language model (LLM) named Helium. It has the ability to speak in a variety of dialects and 70 distinct emotional and speaking styles. Additionally, Moshi has the ability to manage two audio streams concurrently, allowing it to converse and listen simultaneously.
The AI chatbot, named after the manner in which people in Japan answer phone calls, responds faster than GPT-4o’s Advanced Voice Mode, which usually takes 232–320 milliseconds. It only takes 200 milliseconds.
According to Kyutai, the goal was to educate Moshi the different nuances and tones of human speech. The organization even worked with a professional voice artist to improve the voice quality.
According to Kyutai, the chatbot’s objective is to become an open source project, meaning that everyone will have access to the model’s code and framework, allowing users to utilize it securely and without worrying about their privacy.
Although Moshi is quicker than GPT-4o, the company claims that this is simply a research prototype meant to demonstrate the bot’s response time and capacity to mimic not just phrases but also tones and voices.
It turns out that Kyutai is also developing a system for tracking signatures, watermarking, and audio identification using AI that will eventually be linked with Moshi. Furthermore, even while this might not be the ChatGPT rival we have been waiting for, it is undoubtedly a significant advancement in the creation of offline, open-source models.
