Meta CEO Mark Zuckerberg announced the release of AudioCraft.
Facebook parent Meta on Wednesday released the code and model weights for several AI tools designed to create sound from text-based prompts.
Why it matters: The use of generative AI could make it easier for those looking to build video games or digital worlds to add background audio as well as visuals.
Driving the news: Meta released three tools for research use under the AudioCraft moniker.
- MediaGen is designed for creating music from text and was trained using music owned or licensed by Meta.
- AudioGen is designed for creating sound effects based on a written description.
- A third tool, EnCodec, is a decoding engine that Meta is also making available to researchers.
Between the lines: Meta is aiming to boost adoption of its AI tools by making them broadly available, as was the case with last month's release of Llama 2, which is free for commercial use.
Yes, but: Meta acknowledges a lack of diversity in the types of music used to train the system, as well as the use of English-only metadata.