Have a listen for yourself.
Google researchers just showed off the latest tricks they've taught MusicLM, their artificial intelligence model that can turn simple text prompts into fully-fledged music.
And, judging by the samples they've shared so far, which range from 30 seconds to entire five-minute tracks, it's pretty impressive — but not perfect, especially when it comes to non-electronic genres and the simulation of human vocals.
One particularly smooth-sounding track imagines the "main soundtrack of an arcade game," while another interprets "swing" music in an eerie and almost terrifyingly robotic way that might have Duke Ellington turning in his grave.
Humming and Whistling
It may not be the first music-generating tool out there — it's not even the first time we've come across MusicLM — but this latest iteration has some new and powerful tricks up its sleeve.
As detailed in a yet-t0-be-peer-reviewed paper shared by the researchers, this version of MusicLM "can be conditioned on both text and a melody in that it can transform whistled and hummed melodies according to the style described in a text caption."
The results are impressive, as evidenced by the samples provided: a small ten-second hummed or whistled melody is easily transformed into an entire a cappella performance, guitar solo, or string quartet rendition. The AI can even turn a finger-strummed guitar sample into a sultry jazz saxophone passage.
It's an iterative improvement over previous efforts, the researchers argue.
"Our experiments show that MusicLM outperforms previous systems both in audio quality and adherence to the text description," the paper reads, comparing their tool to other competing efforts like Mubert, a music-generating AI.
But before you ask: no, the researchers won't be releasing the tool to the public — for one important reason: the potential for copyright infringement or "potential misappropriation of creative content," something the music industry is already paying close attention to.
READ MORE: Google’s new AI turns text into music [The Verge]
More on AI music: Record Labels Terrified by Rise of AI Music Generators