Meta Releases AI Music Generator Trained To Run “10,000 High-Quality Licensed Music Tracks”.

Photo credit: Technivation

Meta has released its own AI music generator called “MusicGen” which is trained on 10,000 licensed music tracks.

The AI ​​music generator works in a similar way to Google’s MusicLM, generating an approximately 12-second audio snippet based on a text prompt. I experimented with MusicLM’s model when it was first released and found that it produces electronic music and synthwave quite well, but not much else. MusicGen wants to get better in a variety of genres.

MusicGen has been trained with 20,000 hours of music, including 10,000 “premium” licensed tracks and 390,000 instrument-only tracks from ShutterStock and Pond5. While the model itself is open source, Meta did not provide the code it used to train the model. Instead, pre-trained models are available for download. The results of MusicGen and MusicLM will not put musicians out of work any time soon.

In order to request an acceptable chunk of audio from a text-to-audio AI, you need to understand how to describe what you want to hear. Simple prompts like “ambient chiptune music” are so open that simply re-feeding the prompt into the music generator will generate wildly different songs after each generation.

Meanwhile, a prompt like “Slow bass and drum led reggae song. Sustainable electric guitar. High pitched bongos with ringing tones. The vocals are relaxed, have a relaxed feel and are very expressive,” the language model will help create something that sounds very similar after each successful generation. As generative AI advances, these language models are getting better and better at producing sounds that are pleasing to the human ear — albeit a bit soulless.

It also means that the era of deepfake music will become even more difficult to distinguish as these models become more prevalent. We’ve already seen viral titles like “Heart On My Sleeve” dominating social media like TikTok and YouTube. The big three labels are discussing new AI precautions to combat deep-fake music, taking musical concepts to create a five-finger discounted mix of the established sound of popular artists.