Google’s innovative AI tool designed to generate high-quality music based on text prompts and humming, MusicLM, is now available for the public to explore and experiment with.
According to Google, MusicLM operates by simply entering a prompt like “soulful jazz for a dinner party”. The model then produces two distinct versions of the requested song, allowing users to vote for their preferred version and contribute to the improvement of the AI model.
The MusicLM model has been trained on an impressive dataset of five million audio clips, totaling a staggering 280,000 hours of music at 24 kHz.
During the initial launch in January, Google showcased MusicLM's remarkable audio generation capabilities through captivating examples sourced from rich captions. You can listen to these intriguing results right here.
Google emphasises that MusicLM is an experimental tool that can unleash your creativity, regardless of whether you're an established musician or just beginning your musical journey.
The company shared a workshop video to demonstrate MusicLM in action, featuring the collaborations between a sound artist, a Google Arts & Culture Artist in Residence, and a dedicated Google researcher.
As Google's researchers highlight, future developments may focus on lyrics generation, enhancing text conditioning, improving vocal quality, and modeling high-level song structures such as intros, verses, and choruses.
In January, Google not only introduced the tool but also unveiled a research paper, delving into the intricate development process that brought this innovation to life. While it extends the range of tools available to assist humans in creative music tasks, the research paper acknowledges several risks associated with its use and the underlying model.
These risks include the potential reflection of biases present in the training data, which raises questions about the appropriateness of music generation for underrepresented cultures and concerns regarding cultural appropriation.
Furthermore, the paper emphasises the importance of responsible model development, conducting thorough studies on memorisation and employing methodologies that prioritise semantic modeling.
Interestingly, Google's public release of MusicLM coincided with the announcement by CEO Sundar Pichai of a major AI push, featuring a series of AI-powered updates across various Google products, marking an exciting inflection point in their AI journey.
As part of this AI initiative, Google plans to expand the availability of its conversational AI tool and Bard, a Chat GPT rival, to over 180 countries following its successful launch in the UK and US.
Moreover, Google has upgraded Bard to its state-of-the-art language model, PaLM 2, boasting advanced math and reasoning skills, as well as coding capabilities.
The release of MusicLM arrives amidst growing concerns surrounding the use of generative AI in music, particularly regarding copyright infringement. The industry is closely monitoring the impact of AI-generated music, as exemplified by the recent viral version of “Heart on My Sleeve” featuring AI-generated vocals imitating the voices of Drake and The Weeknd.