Meta’s has unveiled its latest open-source ‘AudioCraft’ AI tool that could change the way we create music.
The AI tool will allow professionals and casual users alike, to produce high-quality audio such as music and sound effects using text prompts.
While mentioning some of the use cases, Meta said that AudioCraft will enable professional artists to create compositions without needing to play an instrument and creators will be able to generate soundtracks with ease.
Meta’s AudioCraft AI tool consists of three models: MusicGen, AudioGen and EnCodec. While MusicGen is trained using Meta’s own music library and generates music from text prompts, AudioGen is trained in public sound effects and generates audio from text prompts.
(For top technology news of the day, subscribe to our tech newsletter Today’s Cache)
The EnCodec decoder on the other hand allows higher quality music generation with fewer artifacts.
Meta has also released its pre-trained AudioGen models, that will allow users to generate environmental sounds and sound effects like cars honking or footsteps on a wooden floor.
Meta believes that while AI for images, video, and text has seen a lot of growth and excitement, audio has seemed to lag behind a bit.
Now, Meta is also sharing all of the AudioCraft model weights and code and open-sourcing these models. This will allow researchers and practitioners to build on Meta’s ecosystem and train their own models with their own datasets.
As part of its official blog, Meta said “Generating high-fidelity audio of any kind requires modeling complex signals and patterns at varying scales. Music is arguably the most challenging type of audio to generate as it’s composed of local and long-range patterns, from a suite of notes to a global musical structure with multiple instruments.”
Published - August 03, 2023 12:30 pm IST
