Meta just unveiled an open-source artificial intelligence (AI) code called AudioCraft. It allows users to create music and sounds from simple written requests.
A few weeks after announcing the commercialization of its Llama 2 language model in partnership with Microsoft Azure, Meta continues its breakthrough in the generative AI sector.
Open source models
AudioCraft is made up of three distinct models. The first, MusicGen, was trained on ” 20,000 hours of music owned by Meta or specifically licensed “. It is able to generate music from texts.
The second is called AudioGen. Trained through public sound effects, it creates sounds from written prompts, such as a dog barking or footsteps. The latest model is an improved version of EnCodec, which can generate music from ” better quality with fewer artifacts “.
” While generative AI has generated a lot of excitement for images, video, and text, audio seems to lag a bit. Work is in progress, but it is very complicated and not very open, which makes it difficult for people to rub shoulders with it writes the company in a blog post. She decided to release the AudioCraft code as open source so that ” other researchers can more easily test new approaches to limit or eliminate potential biases and misuse of generative models “.
AI is far from unanimous in the artistic community
Meta isn’t the first tech giant to release a model that generates audio. Last January, Google presented MusicLM, an AI capable of composing music from text.
According to Meta, AudioCraft can spawn a new wave of songs, the same way synthesizers changed music when they first became popular. While AI is welcomed with open arms by some artists, like Grimes, others continue to warn against the dangers of generative AIs for their work. For now, the capabilities of technology in the musical field are still limited, but it is certain that its capabilities will multiply rapidly.
Source : Meta
0