Google has built an AI that can generate music from text descriptions, but won’t export it • TechCrunch

Google’s impressive new AI system can generate music of any genre while providing a text description. But the company, fearing the risks, has no immediate plans to release it.

named MusicLMIt is not certain that Google is the first AI system for songs. There were other attempts, incl Spreadan artificial intelligence that composes music by clocking it, as well Dancing spreadAudioML and Google’s OpenAI jukebox. But due to technical limitations and limited training data, none of them were able to produce songs that were particularly complex in composition or high fidelity.

Perhaps MusicLM will be the first to do so.

detailed in the academy paperMusicLM was trained on a dataset of 280,000 hours of music to learn how to create coherent songs of descriptions of — as the creators put it — “great complexity” (eg, “a charming jazz ballad with memorable saxophone solos and a vocal solo” or “techno”). Berlin in the ’90s with low bass and a powerful kick.”

It’s hard to overstate how Hassan audio samples, given that there are no musicians or instrumentalists in the episode. Even when providing long and somewhat meandering descriptions, MusicLM manages to capture nuances such as tracks, melodies, and moods.

The sample comment below, for example, includes the “induces the experience of being lost in space” part, and it certainly delivers on that front (at least to my ears):

Here’s another sample, built from a description starting with the sentence “Master soundtrack for an arcade game”. Reasonable, right?

MusicLM’s capabilities go beyond producing short clips of songs. Google researchers have shown that the system can build on existing melodies, whether it’s humming, singing, whistling, or playing an instrument. Furthermore, MusicLM can take several descriptions written in sequence (eg, “meditation time”, “wake up time”, “running time”, “100% giving time”) and create a kind of melodic “story” or narrative as long as to several minutes—just right for a movie soundtrack.

See below, which came from the sequence “Electronic song played in a video game”, “Meditation song played by a river”, “Fire”, “Fireworks”.

That’s not to say MusicLM is flawless — far from it, honestly. Some samples have a distorted quality to them, which is an unavoidable side effect of the training process. And while MusicLM can technically create vocals, including choral harmony, it leaves a lot to be desired. Most of the “lyrics” range from barely English to pure nonsense, sung by synthesized voices that sound like an amalgamation of several artists.

However, Google researchers noted the many ethical challenges posed by a system like MusicLM, including the tendency to incorporate copyrighted material from the training data into songs created. During one experiment, they found that about 1% of the music generated by the system was copied directly from the songs it trained on—a threshold apparently high enough to dissuade them from launching MusicLM in its current state.

“We acknowledge the potential misappropriation risks of creative content associated with the use case,” the research co-authors wrote. “We strongly emphasize the need for further work in the future to address these risks associated with music generation.”

Assuming MusicLM or a system like this becomes available one day, it seems inevitable that major legal issues will come to the fore. They already have, albeit about simpler AI systems. In 2020, Jay-Z’s record label filed copyright strikes against a YouTube channel, Vocal Synthesis, for using artificial intelligence to create Jay-Z’s covers of songs like Billy Joel’s “We Didn’t Start the Fire.” After initially removing the videos, YouTube reinstated them, finding that the removal requests were “incomplete”. But deep The music still rests on murky legal ground.

a White papers Written by Eric Sunray, now a legal intern with the Music Publishers Association, he argues that AI-powered music generators like MusicLM infringe the copyright of music by creating a “coherent sonic tapestry from the works they assimilate in training, thus violating copyright law.” US copyright.” After the release of Jukebox, critics also questioned whether training AI models on copyrighted musical material constituted fair use. Similar concerns have been raised about the training data used in image-generating, coding, and text AI systems, which are often deleted from the web without the creators’ knowledge.

From a user perspective, Andy Baio from Waxy speculate That music generated by the AI ​​system will be considered a derivative work, in which case only the original elements will be protected by copyright. Of course, it is not clear what can be considered “original” in such music. Using this music commercially is entering uncharted waters. It’s a simpler matter if the music created is used for purposes protected under fair use, such as parody and commentary, but Baio anticipates that courts will have to make rulings on a case-by-case basis.

It may not be long before there is some clarity on the matter. Several lawsuits Making their way through the courts could potentially have an impact on AI for music generation, including rights over the rights of artists whose work is used to train AI systems without their knowledge or consent. But time will tell.

Leave a Comment