The company AI has released the basic model that empowers Maya, the impressive realistic sound assistant.
The model, which is 1 billion parameters in size (“parameters” that refer to individual model ingredients), is under an apache 2.0 license, which means it can be used commercially with some restrictions. Called CSM-1B, the model generates “Audio RVQ Code” from text and audio inputs, according to Sesame’s description on the AI dev face face platform.
RVQ refers to “quantization of the remaining vector”, a technique for coding the audio in discrete signs called code. The RVQ is used in a number of the latest audio technologies, including Google Soundstream and Meta Encodec.
CSM-1b uses a model from the Meta’s Llama family as the spine paired with a “decipher” audio component. A well -arranged variant of CSM Maya powers, says Sesame.
“The open-source model here is a base of base generation,” writes Sesame in CSM-1B embrace facial warehouses and Github warehouses. “It is capable of producing a variety of voices, but it is not well regulated with any specific voice (…) the model has several non-English language capacities due to data pollution in training data, but it is likely that it will not do well.”
Unciling what sesame data has used to train CSM-1B. The company did not say.
It is worth noting that the model has no real protective measures to speak. SESAME has a system of honor and simply encourages developers and users not to use the model to imitate a person’s voice without their consent, create fraudulent content as fake news, or engage in “harmful” or “malicious” activities.
I tried the demonstration in the facial hug and the cloning of my voice took less than a minute. From there, it was easy to generate a speech at my heart’s wish, including controversial topics such as Russian choices and propaganda.
Consumer reports recently warned that many well -known voice cloning tools with him in the market do not have “significant” safeguards to prevent fraud or abuse.
Sesame, founded by the Oculus co-creator, Iribe, went viral in late February for his Tech assistant, who approaches the cleaning of the UNCANNY valley territory. The other assistant of the top and the sesame, Miles, breathe and speak with ungratefulness and can be interrupted as they speak, just like the way Openai’s voice.
Sesame has raised an undetected amount of capital by Andreessen Horowitz, Spark Capital and Matrix Partners. In addition to building a sound assistant, the company says it is the prototypization of the glasses of it “designed to wear all day” that will be equipped with its custom models.
