Sesame, the startup behind the virtual virtual assistant Maya, releases its AI basic model
You have a business Sesame published the basic model that feeds Maya, the Impressive realistic vocal assistant.
The model, which is 1 billion size parameters (“parameters” referring to the individual components of the model), is under an Apache 2.0 license, which means that it can be used commercially with few restrictions. Called CSM-1B, the model generates “RVQ audio codes” from text and audio inputs, according to The description of Sesame on the AI platform Hurching the face.
RVQ refers to the “quantification of residual vectors”, a technique to code the audio in discreet tokens called codes. RVQ is used In a number of recent AI audio technologiesIncluding Google and Meta encode Soundstream.
CSM-1B uses a model from Meta’s Llama Family Like its backbone associated with an audio component “Decoder”. A refined variant of CSM Powers Maya, says Sesame.
“The open-open model here is a basic generation model,” writes Sesame in CSM-1B Face And Girub References. “He is able to produce a variety of votes, but it has not been refined in a specific voice […] The model has a certain capacity for non -English -speaking languages due to the contamination of data in training data, but that probably won’t do well. »»
We do not know which sesame data used to train CSM-1B. The company has not said.
It should be noted that the model has no real guarantees to speak. Sesame has an honorary system and simply urges developers and users not to use the model to imitate a person’s voice without their consent, create misleading content as false news or engage in “harmful” or “malicious” activities.
I tried The demo By hugging the face, and the cloning of my voice took less than a minute. From there it was easy to generate a discourse in the desire of my heart, including on controversial subjects such as Russian election and propaganda.
SESAME, co-founded by the co-creator of Oculus, Brendan Iribe, became viral at the end of February for his assistant technology, which is close to the Uncanny Valley territory. The other assistant of Maya and the Sesame, Miles, breathes and speaks with the Defluences, and can be interrupted during the Word, A bit like the vocal mode of Openai.
Sesame has raised an unhappy quantity of capital from Andreessen Horowitz, Spark Capital and Matrix Partners. In addition to building Tech assistant vocal technology, the company says it is the prototyping of AI glasses “designed to be worn all day” which will be equipped with its personalized models.
