In the modern age of artificial intelligence, Meta has introduced a groundbreaking application that’s redefining how we create music. The company recently unveiled MusicGen, an AI that has been trained on an impressive 20,000 hours of music and can generate tunes based on human language descriptions.
This revolutionary AI music composer follows the recent launch of Google’s MusicLM, which creates music from text prompts or humming. MusicGen, however, adds its own spin by enabling users to describe the type of music they wish to hear. Whether users crave “a dynamic fusion of hip-hop and orchestral elements, encapsulating the lively spirit of the city,” or simply a “90s rock song with a guitar riff,” MusicGen can cater to it all.
We present MusicGen: A simple and controllable music generation model. MusicGen can be prompted by both text and melody.— Felix Kreuk (@FelixKreuk) June 9, 2023
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
For added customization, users can upload a song to guide MusicGen in crafting the desired content. The AI responds by producing a 12-second music clip in just a matter of minutes.
In Meta’s internal evaluation, MusicGen stood its ground and even outperformed other comparable programs like MusicLM, Diffusion, and Noise2Music. It delivered high-quality samples that were melodically aligned with a given harmonic structure and accurately adhered to the provided textual description.
Meta’s innovative AI music composer was trialed in three versions, each varying in the degree of music detail they provide with 300 million, 1.5 billion, and 3.3 billion parameters. Evaluations revealed that human participants preferred the results from the model with 1.5 billion parameters, resonating with the famous music production mantra, “Less is more.”
Despite the promising potential of MusicGen, some concerns arise in the realm of copyright law. Meta reassures that all songs used in training have been cleared by copyright holders. However, when users input a song or artist’s name in their description, it could potentially infringe upon copyright laws. Furthermore, allowing users to upload a song to influence the final output blurs the lines between original and copied content.
Prominent figures in the music industry have voiced contrasting views on AI music. Australian singer and composer Nick Cave criticizes the concept, asserting that AI lacks the human experiences necessary for true creation. Ben Beaumont-Thomas of The Guardian also chimed in, stating that “AI will always be a tribute act.”
However, not everyone is pessimistic. Spotify’s CEO Daniel Ek sees potential in AI-generated music: “This could be potentially huge for creativity… That should lead to more music [which] we think is great culturally.”
As the AI music landscape continues to evolve, it remains to be seen whether these tools will complement human creativity or spark ongoing debates about originality and copyright. Either way, the music world has undoubtedly entered a new era, and Meta’s MusicGen is leading the way.