Text to Music: Google’s MusicLM Now Accessible to the Public
Today, Google has unveiled MusicLM, an experimental AI tool designed to transform text descriptions into musical compositions. Accessible through the AI Test Kitchen app on the web, Android, or iOS, MusicLM enables users to input prompts such as “soulful jazz for a dinner party” or “create hypnotic industrial techno” and generate multiple versions of the resulting song.
Within MusicLM, users can specify their desired instruments, such as “electronic” or “classical,” as well as the intended vibe, mood, or emotion, allowing for further customization of the generated music.
When Google initially showcased MusicLM in an academic paper earlier this year, it expressed no immediate plans for release. The paper’s coauthors highlighted the ethical challenges associated with such a system, including the potential incorporation of copyrighted material from the training data into the generated compositions.
However, Google has since collaborated with musicians and conducted workshops to explore how this technology can enhance the creative process. As a result, the version of MusicLM available in the AI Test Kitchen does not generate music featuring specific artists or vocals, although the implications of this decision remain open to interpretation.
Nevertheless, the broader challenges surrounding generative music are unlikely to be easily resolved. In 2020, Jay-Z’s record label filed copyright strikes against a YouTube channel named Vocal Synthesis, which used AI to produce Jay-Z covers of songs like Billy Joel’s “We Didn’t Start the Fire.” After initially taking down the videos, YouTube reinstated them, citing incomplete takedown requests.
The legal status of deep-faked music remains uncertain and subject to an ongoing debate.
The rise of homemade tracks utilizing generative AI to replicate familiar sounds, often passing as authentic or near enough, has gained significant attention. Music labels have been quick to notify streaming platforms about these tracks, citing concerns over intellectual property. Unlike the Jay-Z case, where YouTube reinstated the AI-generated covers, Spotify recently removed tens of thousands of AI-generated songs from startup Boomy following a complaint from Universal Music Group.
Eric Sunray, currently a legal intern at the Music Publishers Association, outlined in a whitepaper that AI music generators like MusicLM infringe upon music copyright. Sunray argues that these AI systems create coherent audio by drawing from existing works during training, thereby violating the reproduction right specified in the United States Copyright Act. The fact that AI, such as MusicLM, “learns” from existing music to produce similar effects raises concerns among some artists.
The ongoing lawsuits related to music-generating AI may eventually bring some clarity to the matter. These legal battles will likely address the rights of artists whose works are used to train AI systems without their knowledge or consent.
Only time will reveal the outcome of these developments and provide a clearer understanding of the implications surrounding music-generating AI.
