Meta launched AudioCraft on Wednesday, an AI tool that lets users create music and sounds from scratch with text-based inputs as part of CEO Mark Zuckerberg’s bigger push into generative AI.
Mark Zuckerberg's Meta is doing a big push with AI tools, launching AudioCraft on Wednesday, after ... [+]
Meta presented open-source AudioCraft as a tool that aims to make generative AI for audio simpler and more broadly available.
All Facebook users can install AudioCraft and Meta is particularly inviting researchers and music professionals to use it.
The new tool is the result of a combination of three AI-generative models: MusicGen takes text inputs to create music, AudioGen does the same with sounds such as footsteps or barking dogs and an improved version of EnCodec helps improve the perceived quality of the two earlier.
MusicGen was trained on 20,000 hours of licensed music owned by Meta or licensed specifically for this purpose, while AudioGen was trained on public sound effects.
Meta introduced its first version of EnCodec as an AI tool to compress and decompress audio files without losing quality for quick and easy sharing in October 2022. It was aimed at enhancing the quality of all sounds, not only music. At the time, it particularly targeted voice calls and voice messages, especially under adversary circumstances such as poor network connectivity. The model has evolved since and is today introduced together with AudioGen and SoundGen as a tool that will help synthetic sounds and music seem real when they are not.
While some artists have embraced AI-generative tools to get more creative, others have been critical of copyright infringements. Drake and the Weeknd weren’t happy to get mimicked in the TikTok viral hit “Heart on My Sleeve,” a track that claimed to use AI versions of their voices to create a passable deepfake. On the flip side, Paul McCartney is among the more enthusiastic. He recently announced a Beatles song from 1969 will be finished with AI help.