You have a company Sesame released the basic model that drives Maya, Impressively realistic voice assistant.
The model, which has 1 billion size parameters (“parameters” regarding individual elements of the model), is under the APACHE 2.0 license, which implies that it may well be used with a small number of restrictions. Called CSM-1B, the model generates “RVQ Audio Codes” from text and audio texts, based on Sesame description on the AI Hulging Face Developer Platform.
RVQ refers to “residual vector quantization”, sound coding techniques with discrete tokens called codes. RVQ is used In many of the latest AI Audio technologiesincluding Google Soundstream and Meta Encodec.
CSM-1B uses a model from the Lama Meta family as a spine in combination with the “decoder” component. Sesame says the refined CSM Maya Maya variant.
“The model here is Open Sourced is a base generation model,” writes Sesame in CSM-1B Hugging AND Girub Repositories. “He is able to produce various voices, but has not been refined with any specific voice […] The model has a certain ability to non -English languages due to data pollution in training data, but it probably won’t be good. “
It is not clear what data sesame used to coach CSM-1B. The company didn’t say.
It is value noting that the model has no real security. Sesame has an honorary system and simply encourages programmers and users not to make use of the model to mimic the voice of a person without their consent, creating misleading content, resembling false messages or are involved in “harmful” or “malicious” activities.
I attempted Demo It took lower than a minute to hug my face and cloning my voice. From there, it was easy to generate speech with the desire of my heart, including in controversial topics, resembling elections and Russian propaganda.
Consumer reports have recently warned that many popular tools for cloning voice driven by artificial intelligence on the market I have no “significant” security To prevent fraud or abuse.
Sesame, co -founded by the co -creator of Oculus, Brendan Ilim, at the end of February became popular for a technology assistant, which is much like the purification of the territory of Uncanny Valley. The second assistant of Maya and Sesame, Miles, take breath and discuss with disappointments, and could be interrupted while speaking, as did OpenAi’s voice mode.
Sesame collected an undisclosed amount of capital from Andreessen Horowitz, Spark Capital and Matrix Partners. In addition to building a voice assistant, the company claims that it prototypes AI glasses “designed to be worn all day”, which will likely be equipped with non -standard models.