Ai Company Sesame has launched the base model Maya, the Impressively realistic voice assistant.
The model, which has a size of 1 billion parameters (“parameters” that refer 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 entries, according to The description of Sesame on the AI Dev Hugging Face platform.
RVQ refers to the “quantization of the residual vector”, a technique to encode the audio in discrete tokens called codes. RVQ is used In several recent AU audio technologiesincluding Soundstream and Google MetaC.
CSM-1B uses a family flame family model as its spine combined with a “audio” decoder “component. An adjusted variant of the Mayan CSM powers, says Sesame.
“The open source model here is a base generation model,” Sesame writes in CSM-1b’s Hugged face and GITHUB Repositories “He is able to produce a variety of voices, but he has not adjusted in any specific voice […] The model has some capacity for languages that are not English due to data pollution in training data, but it is likely not to work well. “
It is not clear what data sesame used to train CSM-1B. The company did not say.
It is worth noting that the model has no real safeguards to talk about. Sesame has an honor system and simply urges developers and users not to use the model to imitate the voice of a person without their consent, create misleading content as false news or participate in “harmful” or “malicious” activities.
I tried The demonstration On the hugged face, and cloning my voice took less than a minute. From there, it was easy to generate a speech about the desire of my heart, even in controversial issues such as Russian elections and propaganda.
Consumer Reports recently warned that many popular voice cloning tools with AI in the market I have no “significant” safeguards to prevent fraud or abuse.
Sesame, co -founded by Oculus’s cooker, Brendan Iribe, went viral at the end of February because of its assistant technology, which approaches the cleaning of the territory of Valle. The other Maya and Sesame assistant, thousands, breathes and talks with dysfluences, and can be interrupted while speaking, as well as OpenAi’s voice mode.
Sesame has increased an unrelated amount of capital of Andreessen Horowitz, Spark Capital and Matrix Partners. In addition to building voice assistant technology, the company says that it is prototypes of “” designed to use all day “glasses that will be equipped with their personalized models.