
Sesame, the startup behind the viral virtual assistant Maya, open-sources its base AI model
Sesame, the company responsible for the incredibly realistic voice assistant Maya, has made a significant move by releasing the foundation of its technology as an open-source model. The base AI model, dubbed CSM-1B, boasts an impressive 1 billion parameters and is available under an Apache 2.0 license.
According to reports, the CSM-1B model generates “RVQ audio codes” from text and audio inputs, a technique employed in recent AI audio technologies such as Google’s SoundStream and Meta’s Encodec. The model uses a backbone derived from Meta’s Llama family alongside an audio “decoder” component. A fine-tuned variant of CSM powers Maya, the company claims.
Sesame notes that this open-sourced model is capable of producing various voices but has not been trained on any specific voice. Additionally, it can recognize some non-English languages due to data contamination in the training set, although performance may vary.
The company did not disclose the data used for training the model.