Sesame has released CSM-1B, a 1 billion parameter AI model that generates audio codes from text and audio inputs. This model utilizes residual vector quantization (RVQ) for encoding, similar to technologies used by Google and Meta. While the base model is open-sourced and commercially usable, it lacks fine-tuning for specific voices and has limited capacity for non-English languages. Sesame has an honor system request for ethical use, discouraging misuse such as impersonation or creating fake news. The company, which gained attention from its AI assistant Maya, leaves developers to navigate the proper usage of this technology.
The model, called CSM-1B, generates "RVQ audio codes" from text and audio inputs, demonstrating a significant advancement in AI voice generation technology.
Sesame urges developers to use the model responsibly, warning against creating misleading content or mimicking voices without consent.
Collection
[
|
...
]