Meta's AI Music Generator Goes Open-Source

In a bid to keep up with its competitors, Meta has unveiled its own AI-powered music generator, and what sets it apart is its commitment to open-source technology. Dubbed MusicGen, this innovative tool can transform a simple text description into a 12-second audio clip, capturing the essence of various music styles and genres. With the option to incorporate reference audio, MusicGen aims to blend creativity and melody seamlessly.

By training on an extensive library of 20,000 hours of music, which includes licensed tracks and instrument-only compositions, Meta has equipped MusicGen with a diverse range of musical knowledge. While the company hasn't shared the training code, they have made pre-trained models available for users with compatible hardware to explore.

But how well does MusicGen perform? While it's certainly impressive, it falls short of replacing human musicians. Its output, although melodic and comparable to other AI music generators like Google's MusicLM, still lacks the finesse and artistry of human-made compositions. Nonetheless, MusicGen manages to deliver reasonably pleasing results, even surprising listeners with its ability to generate cohesive tracks, such as jazzy elevator music or lo-fi electro chill beats.

To showcase its capabilities, MusicGen was put to the test against MusicLM, tackling more complex prompts like "lo-fi slow BPM electro chill with organic samples." Surprisingly, MusicGen surpassed MusicLM in terms of musical coherence, producing tunes that could easily find a place in the trendy world of Lofi Girl.

Notably, MusicGen has embraced a more lenient approach compared to its counterparts. While Google implemented filters to address copyright concerns, Meta has chosen not to restrict MusicGen's output. This decision opens up possibilities for users to experiment with prompts that mention specific artists or styles, although the results may vary.

While generative music continues to evolve, ethical and legal challenges loom large. Artists and rights holders express concerns about AI systems learning from existing music and potentially infringing upon intellectual property. The rise of homemade tracks that utilize generative AI, mimicking familiar sounds, raises questions about authenticity and copyright infringement. With ongoing lawsuits shaping the landscape of music-generating AI, including disputes surrounding artists' rights and the use of their work without consent, clarity in this domain is eagerly awaited.

Meta, in its stance, refrains from imposing usage restrictions on MusicGen and assures that all the music used for training was covered by legal agreements with the right holders. As the discussion surrounding generative AI and its impact on the music industry continues, Meta's MusicGen serves as another step forward in the realm of AI-generated creativity.