On August 2, Meta rolled out a new generative AI tool, AudioCraft, aimed at audio and music creation. This tool enables users to generate music and audio based on text prompts.
AudioCraft integrates three models or technologies-AudioGen, EnCodec, and MusicGen-allowing the generation of high-quality, almost human-like audio and music from textual content. MusicGen, which has undergone training with Meta's proprietary and specially authorized music, can generate music from text prompts. AudioGen, trained on public sound effects, can generate audio from text prompts, such as mimicking a dog barking or footsteps. Coupled with an improved version of the EnCodec codec, users can more efficiently generate higher-quality music.
Meta asserts that the AudioCraft suite of models can produce high-quality audio with long-term consistency and is user-friendly:
"Compared to previous work in this field, we've simplified the overall design of the audio generation model with AudioCraft," the company stated. "We're providing people with a comprehensive method using the existing models Meta has developed over the past few years, while also enabling them to push boundaries and develop their own models."
Meta pointed out that AudioCraft is suitable for the compression and generation of music, sound, and audio files. As it's easy to construct and reuse, those who wish to build better sound generators, compression algorithms, or music generators can do it all within the same codebase, building on the work of others.
"Having a solid open-source foundation will foster innovation and supplement how we produce and listen to audio and music in the future," Meta explained. "With more control, we believe MusicGen can become a new type of instrument, much like when synthesizers first emerged."
All Facebook users can install AudioCraft, with Meta particularly inviting researchers and music professionals to use the tool:
"We view the AudioCraft suite of models as a tool for musicians and sound designers to gain inspiration, helping people brainstorm quickly and iterate their work in new ways. We can't wait to see what people create with AudioCraft."
Meta launched its first version of EnCodec in October 2022 as an AI tool used to compress and decompress audio files without losing sound quality, enabling users to quickly and easily share audio documents. Its goal was to enhance the quality of all audio files, not just music files. At that time, it was specifically aimed at improving the quality of voice calls and voice messages, especially under unfavorable conditions like poor network connectivity. The model has since evolved, and now, in combination with AudioGen and SoundGen, it serves as a tool that makes synthesized sounds and music sound more realistic when played.
While some artists have adopted artificial intelligence generation tools to gain more creativity, other artists have been critical, citing concerns about potential copyright infringements.
