Meta Launches ChatGPT for Music Competing with Google MusicLM

Meta launched an open-source deep learning language model based on Artificial Intelligence (AI), MusicGen. (photo: dock.pexs)

JAKARTA - Not wanting to be left behind by Google, Meta launched an open source deep learning language model based on Artificial Intelligence (AI), MusicGen.

Last month, Google released a similar music generator called MusicLM, but MusicGen seems to produce slightly better results.

Developed by the Audiocraft team at Meta, MusicGen is like a music version of ChatGPT in that it can generate new music on text requests and can sync with existing songs.

We present MusicGen: A simple and controllable music generation model. MusicGen can be prompted by both text and melody.

We release code (MIT) and models (CC-BY NC) for open research, reproducibility, and for the music community: https://t.co/OkYjL4xDN7 pic.twitter.com/h1l4LGzYgf

— Felix Kreuk (@FelixKreuk) June 9, 2023

Users just need to enter a short text description of the type of music they want to hear and in no time, the AI ​​will create a 12-second track according to the instructions.

For example, one could tell MusicGen to produce a track "lofi slow BPM electro chill with organic samples", and the resulting audio sounds like something one would hear on YouTube Lofi Girl radio.

Meta states, the Audiocraft team used 20.000 hours of licensed music for training, including 10.000 high-quality tracks from internal datasets, along with Shutterstock and Pond5 tracks.

To make it even faster, they used the company's EnCodec 32Khz audio tokenizer to output smaller pieces of music that could be processed in parallel.

However, unlike MusicLM, MusicGen cannot perform vocals, only instrumentals. Currently, the new AI Meta model is available for free on the Hugging Face website, as quoted from Engadget, Tuesday, June 13th.

"Unlike existing methods like MusicLM (Google), MusicGen does not require self-supervised semantic representation (and) only has 50 automatic regression steps per second of audio," tweeted Hugging Face ML Engineer Ahsen Khaliq.