Meta has launched its own artificially intelligent music generator to rival Google’s.
Meta’s MusicGen technology creates music from words. Based on demonstrations, this can transform text descriptions into approximately 12 seconds of sounds.
Felix Kreuk, an AI researcher for Meta AI, announced this development via his Twitter handle.
Meta reports that MusicGen received training on over 20,000 hours of music — 10000 “high-quality” licensed files as well as 390,000 instrument recordings from Pond5 and ShutterStock stock libraries –
MusicGen is an easily configurable music generation model which responds to text as well as melody for prompting music creation.
“We release code (MIT) and models (CC-BY NC) for open research, reproducibility and the music community – Felix tweeted.”
The company has provided pre-trained models that anyone equipped with suitable hardware, specifically GPUs with 16GB or more memory, can run. Unfortunately, however, no code used to train these models was made public by them.