Meta introduces text-to-music generator AudioCraft

Facebook parent company Meta has unveiled an open source music generator that lets users generate music and sounds through text prompts. AudioCraft can also be used to compress audio.

AudioCraft consists of three AI models. MusicGen can be used to turn text input into music. According to Meta, the model has been trained on ‘20,000 hours of music’. This concerns music owned by Meta and licensed material. The AudioGen model is able to create sounds based on prompts, such as barking dogs or footsteps. This model is trained on public sound effects. In addition, the tech company has improved the existing EnCodec model. The decoder can be used to convert music to a higher quality.

Google released a similar language model at the beginning of this year, called MusicLM. This model is not publicly accessible. This resulted in, among other things, a song with a voice similar to that of Drake and The Weeknd. The AI-generated song Heart On My Sleeve was quickly removed from streaming services and also raises the question to what extent such an AI song violates the copyright of musicians.