As GenAI tools begin to transform the music industry in incredible – and in some cases ethically problematic – ways, Google is accelerating its investments in AI technology to create new songs and lyrics.
The search giant today unveiled MusicFX, an upgrade to MusicLM, the music generation tool that Google launched last year. MusicFX can create songs up to 70 seconds long and music loops, delivering what Google claims is “higher quality” and “faster” music generation.
MusicFX is available in Google AI Test Kitchen, an app that lets users try out experimental AI-based systems in the company’s labs. Technically, MusicFX was launched for select users in December – but now it’s generally available.
Image credits: Google
And it’s not great, I must say.
Like its predecessor, MusicFX allows users to enter a text prompt (“two nylon-string guitars playing in flamenco style”) to describe the song they want to create. The tool generates two versions of 30 seconds by default, with options to lengthen the tracks (to 50 or 70 seconds) or automatically stitch the start and end to loop them.
A new addition is suggestions for alternative descriptor words in prompts. For example, if you type in “country style,” you might see a drop-down list with genres such as “rockabilly style” and “bluegrass style.” For the word “catchy,” the dropdown might contain “chill” and “melodic.”
Image credits: Google
Below the prompt field, MusicFX provides a word cloud of additional recommendations for relevant descriptions, instruments and tempos to add (e.g. “avant-garde”, “fast”, “exciting”, “808 drums” ).
So how are you ? Well, in my brief testing, the MusicFX samples were…good? Truth be told, music generation tools are getting to a point where it’s difficult for this writer to distinguish between releases. The current state of the art produces incredibly clear, crisp-sounding tracks – but tracks tending towards the boring, uninspired and melodically unfocused.
Maybe it’s the SAD It gets to me, but one of the prompts I followed was “a house music song with funky, danceable, uplifting beats, with summer rooftop vibes.” MusicFX delivered, and the songs weren’t bad – but I can’t say they come close to the best DJ sets I’ve heard recently.
Listen for yourself:
Anything that uses stringed instruments sounds worse, like a cheap MIDI sample, perhaps reflecting MusicFX’s limited training set. Here are two pieces generated with the prompt “a moving melody played on stringed instruments, orchestral, with a strong melodic core”:
And for a change of pace, here’s MusicFX’s rendition of a “slow-tempo, melancholy, mournful guitar song on a moonlit night (sic).” (Forgive the spelling mistake.)
There are some things MusicFX won’t generate – and this cannot be removed from generated tracks. To avoid copyright infringement, Google filters prompts that mention specific artists or include vocals. And it uses Synthesizer IDan inaudible watermarking technology developed by its division DeepMind, to clearly indicate which tracks come from MusicFX.
I’m not sure what type of master list Google uses to filter artists and song names, but I didn’t find it that difficult to defeat. While MusicFX refused to generate songs in the style of SZA and the Beatles, fortunately they had to quickly reference Lake Street Dive – even if the tracks weren’t very well written, I will say.
Speech generation
Google has released a new lyrics generation tool, TextFX, in AI Test Kitchen, designed as a sort of companion to MusicFX. Like MusicFX, TextFX has been available to a small cohort of users for some time, but is now more widely available and improved in terms of “user experience and navigation,” Google says.
As Google explains in the AI Test Kitchen app, TextFX was created in collaboration with Lupe Fiasco, the rap artist and record producer. It is powered by PaLM2one of Google’s text generation AI models, and “(draws) inspiration from the lyrical and linguistic techniques that (Fiasco) has developed throughout his career.”
Image credits: Google
This reporter expected TextFX to be a more or less automated speech generator. But it is certainly not that. Instead, TextFX is a suite of modules designed to facilitate the process of writing lyrics, including a module that searches for words in a category starting with a chosen letter and a module that finds similarities between two unrelated items.
Image credits: Google
TextFX takes some getting used to. But I can see it becoming a useful resource for lyricists – and writers in general, frankly.
You will, however, want to take a close look at its results. Google warns that TextFX “may display inaccurate information, including about people,” and I actually managed to get it to suggest that climate change “is a hoax perpetrated by the Chinese government to harm American businesses.” . Yeah.
Image credits: Google
Questions remain
With MusicFX and TextFX, Google signals that it is investing heavily in GenAI music technology. But I wonder if his concern for keeping up with THE The Joneses rather than addressing the difficult questions surrounding GenAI music, this will ultimately serve it well.
Increasingly, homemade tracks that use GenAI to evoke familiar sounds and voices that can be considered authentic, or at least close enough, are going viral. Music labels were quick to flag AI-generated tracks to streaming partners like Spotify and SoundCloud, citing intellectual property concerns. They were generally victorious. But it remains unclear whether “deepfake” music violates the copyrights of artists, labels and other rights holders.
A federal judge ruled in August that AI-generated art cannot be copyrighted. However, the US Copyright Office has not yet taken a position, only recently begin soliciting public input on AI-related copyright issues. It’s also unclear whether users might find themselves guilty of violating copyright law if they attempt to commercialize music created in the style of another artist.
Google is trying to chart a cautious path toward rolling out GenAI music tools to the YouTube side of its business, which tests AI models created by DeepMind in partnership with artists like Alec Benjamin, Charlie Puth, Charli XCX, Demi Lovato, John Legend, Sia and T-Pain. That’s more than can be said for some of the tech giant’s GenAI competitors, like Stability AIwhich considers that “fair use” justifies training on content without the authorization of the creator.
But with labels pursue GenAI suppliers on copyrighted speech in training data and artists express their discontentGoogle has its work cut out for it – and it’s not letting this inconvenient fact slow it down.
