Meta releases AudioCraft AI sound and song generator and opens its source code
Meta has released AudioCraft, an open-source AI tool that allows generating music and sounds based on text prompts.
AudioCraft consists of three AI models, each of which solves different sound generation tasks:
- MusicGen creates music. It was trained on 20,000 hours of music owned by Meta or licensed specifically for this purpose;
- AudioGen generates sounds by imitating barking dogs or footsteps. The model was trained on recordings collected in public places;
- EnCodec decoder allows you to create sounds with fewer artifacts.
Meta believes that AudioCraft is capable of sparking a new wave of songs, just as synthesizers once changed music. The company also opened the source code of the tool to diversify the data used for its training.
You can listen to samples of music and sounds created by AudioCraft on the project’s website. The source code of the tool is available on GitHub.
Source
ai.meta