Tech Desk 
Mar 17, 2025
Amid an ever-intense race among popular artificial intelligence (AI) startups, Sesame AI, a relatively newer US-based voice AI firm, has publicly released the AI model powering its immensely popular voice assistant Maya.
Sesame AI rolled out CSM-1B, a meticulously designed speech generation model equipped with 1 billion parameters, open-source on March 13, according to AutoGPT.
It's worth mentioning that Sesame AI's voice generation model is freely available with an Apache 2.0 license.
The chief property of the voice AI assistant is its ability to create realistic speech in response to both text and audio inputs.
It uses advanced residual vector quantization (RVQ) technology to produce remarkable lifelike voice outputs, imitating Google’s SoundStream and Meta Encodec.
The resemblance gets the nod of appreciation with Maya being built on Meta’s Llama AI, which allows it to compose a variety of voices without requiring much optimisation.
Although the CSM-1B is now accessible with an Apache 2.0 license to developers willing to leverage its voice generation prowess, it also brings a bunch of limitations for commercial applications.
Nevertheless, Sesame AI's voice AI model has paved the foundations of a spate of innovation in the realm of voice AI assistants.
