HomeAI News
Meta releases AudioCraft, an open source AI tool that allows users to create music, audio through text prompts
107

Meta releases AudioCraft, an open source AI tool that allows users to create music, audio through text prompts

Hayo News
Hayo News
August 3rd, 2023
View OriginalTranslated by Google

On August 2, EST, Meta open sourced a generative AI tool, AudioCraft, that helps users create music and audio through text prompts.

According to Meta's official introduction, AudioCraft includes three core components :

  • MusicGen: Trained using Meta-owned/specially licensed music, generates music from text prompts.
  • AudioGen: Use public sound effects for training to generate audio or extend existing audio to later generate ambient sound effects (e.g. dog barking, car horn, footsteps on wooden floors).
  • EnCodec (improved): A neural network-based audio compression codec that generates higher-quality music with reduced artifacts, or lossless compression of audio files.

After the tool is open-sourced, relevant researchers and practitioners can use their own datasets to train models. The official claims that the AudioCraft series of models can generate high-quality audio stably for a long time, and are easy to use, can provide musicians and sound designers with "inspiration", helping them quickly brainstorm and "iterate their works in new ways".

According to previous reports, in June this year, Meta open sourced the AI ​​​​language model MusicGen, which is based on the Transformer model launched by Google in 2017 . As the model name suggests, MusicGen is mainly used for music generation, which can convert text and existing melodies into complete compositions.

Reference link:

AudioCraft project address: https://github.com/facebookresearch/audiocraft

Reprinted from Emilia DavidView Original

Comments

no dataCoffee time! Feel free to comment