If you have played around with ChatGPT at all, you know how impressive (and terrifying) AI can be. ChatGPT can generate almost anything you ask of it, with prompts as simple or as complex as you can imagine. Well, Google figured out how to do the same. But instead of spitting out poems or computer programs, Google’s AI is writing music and you can visualize its creations at the moment.
Google’s AI, called MusicLM, is designed to produce several minutes of high-quality music based on text prompts. While AI music isn’t necessarily new, the company claims its program produces songs that are more accurate to the original line of lyrics and are of better quality than previous models. Big claims, of course. But, based on these visualizationsthey might just justify them.
The first batch demonstrates how MusicLM can generate music from rich captions, similar to how OpenAI’s ChatGPT generates its famous responses from user queries. Google may provide MusicLM with a caption such as “The main soundtrack of an arcade game. It’s fast and upbeat, with a catchy electric guitar riff. The music is repetitive and easy to remember, but with unexpected sounds like cymbals hitting or drumming,” and the AI will process a song that meets those criteria.
My personal favorite from this section is the track generated by ‘Funky track with a strong, danceable beat and a punchy bass. A catchy keyboard melody adds a layer of richness and complexity to the song” on page two that sounds like something out of the blue Stardew Valley.
Where mysterious valley really however, it occurs whenever Google asks MusicLM to produce vocals. Many of Google’s examples here have AI vocals, and while some of them sound “computer” for lack of a better word, others sound too close to home. When you’re listening with the full knowledge that the voice is 100% artificial, it’s a bit disconcerting. Google even asked MusicLM to generate a rap song, and frankly, the “rapping” is eerily realistic, even if none of the words they’re rapping are real.
G/O Media may receive a commission
health
The Lyma supplement
10 ingredients. A super supplement.
Lyma makes the next generation of supplements – peer-reviewed and formulated to unlock your potential.
If you really want to freak out, scroll down to “Conditioning text and melody.” Here you can hear a series of voices and instruments doing their best with different melodies, and the results are amazing. At the very first option, MusicLM hums “Bella Ciao” and sounds way too good Side note: The Tribal Drums and Flute version gives White Lotus.
We also have the ‘Long Generation’, which creates five-minute songs from brief descriptions such as ‘melodic techno’ or ‘relaxing jazz’. But things get really interesting with Story Mode. Google feeds MusicML multiple captions and the AI adjusts the song based on the given caption. It starts with 15 seconds based on “meditation time” before the caption switches to “wake up time”. The music smoothly changes into a melody, as if starting a new verse, with some creepy vocals to boot.
Creepy vocals aside, it’s fascinating to see the AI change the song without having to change the song. For the most part, it incorporates the new lyrics into the song without feeling like you’re jumping into a new tune. It’s particularly impressive with the third example, where the AI jumps from “pop song” to “rock song” to “death metal song”. While a “rock song” doesn’t really sound like rock, a “death metal song” is just fantastic. I imagine AI Death will be a big hit with metal heads.
Another interesting experiment Google did was to feed MusicLM descriptions of famous paintings to see what kind of music it would produce. If you’ve ever wanted to know what Vincent Van Gogh’s Starry Night would sound like as a song, well, now’s your chance. It’s also fascinating to click through how MusicLM generates specific instrument and genre sounds. The electric guitar is on point, as is the ‘British Indie Rock’, complete with a dramatic twist at the end of the preview.
There’s more to explore from this limited preview of MusicLM, from “Accordion Solo” to comparing different results from the same prompt. Google also published a 15 page research paper on MusicLM if you are interested in reading the technical details of the system. While I hope the company releases a beta version for the public to try, seeing these samples is the next best thing.
[The Verge]