The AI-powered song generator ‘MusicGen’ has been released by Meta.
Meta (formerly Facebook) has unveiled MusicGen, an AI-powered music generator that can transform text descriptions and melodies into audio. The company has made the code and models of MusicGen openly accessible for research purposes and to foster collaboration within the music community.
According to Felix Kreuk, a Research Engineer at Meta AI research, MusicGen is a single-stage transformerLM built upon the EnCodec audio tokenizer. It eliminates the need for multiple models by utilizing efficient token interleaving patterns.
MusicGen was trained on a dataset comprising 20,000 hours of music, including 10,000 licensed tracks and 390,000 instrument-only tracks sourced from ShutterStock and Pond5 stock media libraries.
While Meta’s MusicGen is a recent addition to the AI-powered music generation space, it is worth noting that Google had previously released its own tool called MusicLM. Google’s MusicLM, which can generate high-fidelity music across various genres based on text descriptions, was introduced in January and made available to the public in May.
The AI tool can be accessed through the AI Test Kitchen app on web, Android, and iOS platforms.
