Meta has introduced a groundbreaking generative artificial intelligence tool, capable of translating text prompts into musical compositions. Alongside this announcement they confirmed they are releasing it as open-source for research pursuits. Named AudioCraft, the tool utilises a trio of models – MusicGen, AudioGen, and EnCodec.
In order to teach MusicGen its capabilities, Meta employed a staggering 20,000 hours of licensed music for training, comprising an impressive collection of 10,000 “high-quality” licensed music tracks.
It is noteworthy that during the unveiling, Meta’s researchers acknowledged the ethical issues they confronted while crafting their generative AI models. It’s hard to make a sound judgement on how quickly AudioCraft will affect life as an independent label, but its potential to completely change how music is created, licensed and shared seems inevitable.