Sesame, starting at the end of virual vurual assistant, release ai base model

Sesame, the company AI behind the real voice assistant is real, has released a AI AI model that the virtual power, he recently promised. The model, is 1 billion parameters (“parameters” show individual components of the model), in Apache license 2.0, which means commercial with some limitations. Called CSM-1B, the model returns the “RVQ Audio Code” from text input and audio, according to the sesame input on the face of the AIB platform platform. RVQ shows the “Quantalization Vector Residy,” Engineering for audio token the discussion called code. RVQ is used in many new AI audio technology, including encodec meta and meta. CSM-1B uses the model of the meta family as the backbone paired with the component of the “Decoder” audio. The very good variant of the CSM Powers maya, Sesame says. “Open-source models here are basic generation models,” write fragments in the face that CSM-1B and Repositorial Github. “Can produce multiple sounds, but it has not been attracted by a specific sound […] The model has several languages ​​for non-English language due to data contamination in training data, but it is likely to be done. “It is not clear whether the Sesame data is used to train the CSM-1B. The company has none of the dangerous, and the” scenery, so it is very easy to produce your heart for my heart, including on the topic controversials such as the Russian election and propaganda, are close to the Capital of Opening Voice. In addition to the tech assistant tech area, the company says prototiesping glasses “designed to be used a day “that will be equipped with custom model.

Source link