Hello everyone! Here is Wei Wei Lai Program Life. Today I bring you an open source project of Text2Music - Audiocraft.
Recently, Meta company released a PyTorch library - Audiocraft on the open source website Github. It is a deep learning based audio processing and generation library. At the same time, Meta company also released an artificial intelligence music model based on Audiocraft called MusicGen.
MusicGen is described as "a simple and controllable language model for music generation". You can submit it not only by giving it a textual description of the music you wish to create, but also by giving it a reference audio content, and it will generate a 12-second sample of the music in response.
MusicGen is a single-stage autoregressive Transformer model. MusicGen's training data comes from 20,000 hours of licensed music. It relies on an internal dataset of 10,000 high-quality music tracks, as well as data from the ShutterStock and Pond5 music libraries.
Since the product has just been launched, there are many users, and the audio generation queue and time are long, and the stability of the Demo service is not particularly good, so you need to try it a few times patiently.
MusicGen Sample Demo
Example 1: Prompt word + reference audio
An 80s driving pop song with heavy drums and synth pads in the background
Reference Audio: Reference Audio
Generate audio:
tmprvz9wf9l
Example 2: Prompt Words
90s rock song with electric guitar and heavy drums
generate audio
tmpyjal70ix
Project address : https://github.com/facebookresearch/audiocraft
Demo address : https://huggingface.co/spaces/facebook/MusicGen
If my article is helpful to you, I hope you can share it with more users. Every one of your likes and collections is the source of my continuous creation. If you encounter any problems in use, please feel free to contact me. Thank you for watching, thank you! !