brettworks

I recently listened to Mirage FM, a recording by English musician and visual artist patten (Damian Roach) made using sounds from Riffusion. Riffusion is a website that generates audio based on user prompts, adapting Stable Diffusion‘s text-to-image generator to reconstruct music from a database of spectrograms. (Spectrograms are two-dimentional representations of audio’s frequency, amplitude, and time aspects.) To create Mirage FM, patten began by typing various prompts and recording Riffusion’s output. He then edited and processed the audio into brief tracks (the longest of which is just two and a half minutes). My favorite track is “Drivetime”

Nodding to both the DJ practice of mining old records for sample material and AI’s theoretical concepts, patten describes his workflow as “crate-digging in latent space.” He also acknowledges the hit and miss aspect of collaborating with AI because of the unevenness of Riffusion’s output:

“A lot of what was being spat out by Riffusion was not great. You had to wait a long time for something, a tiny little thing that made you say ‘yes, okay, there’s something there’. A lot of it was looking for things that had a simultaneous sort of familiarity and also a strangeness to them, that I could kind of tease out and push in the direction I wanted to.”

As I listened to Mirage FM I could hear the music two ways. On the one hand, the tracks have a striking, lo-fi mono sound that doesn’t evoke any music I’ve heard before. They aren’t lush or groovy or timbrally attractive. Critic Chal Ravens sums up the sound well: they sound both “familiar and unintelligible.” The music challenges you to try to understand its peculiar artifacts, its cyborg voice traces, and its wonky beats that never settle. Can you imagine the world where the sounds are coming from? On the other hand, the tracks are sort of musically malfunctioning. Maybe the reason is they lack sufficient traces of the human: it’s hard to discern a hand shaping them and a social context in which they could thrive. Mirage FM is both thought provoking and offers little to relate to.

After listening I went to the Riffusion website and typed in some prompts, including “gong”, “ambient Afrobeat”, “techno gamelan”, and “disco marimba.” The results were abysmal, although so too were my prompts. (What were patten’s prompts?) Part of the problem is that different prompts produce barely different results, and no matter the prompt, Riffusion’s output is deeply unmusical. This got me thinking about the various qualities that consitute “being musical.” If we consider, as wt Robina does in his book, Technical Manifesto For The Deviant Sound Engineer, that sound–and by extension, music–“is the phenomenal representation of spirit” (65) then Riffusion’s AI-powered output is, so far, failing at being musical.

But I like that patten made something with Riffusion’s riffs. He explains that his motivation for making a recording based on Riffusion’s output was to “interrogate the way things are” in contemporary music production, and more broadly, to foreground questions about human creativity: “How do our minds work? What is creativity? How are ideas made?” In this regard, Mirage FM succeeds, though, ironically, its success may be due to the finesse it lacks. The ugliness of Riffusion’s sounds might require considerably more work to transform them into something more listenable. While AI-based technologies may mark a turning point in how artists create (though the jury’s still out on that), the works we recognize as creative are so because of the craft that went into their making. So one question about Mirage FM is whether its concept overshadows its sonic reality. The tracks get me thinking, but I’m not moved when I listen. Then again, maybe liking or not liking music is beside the point of such a thought-provoking project?

To close, as I thought more about this music I remembered a 2018 Autechre piece called “32a_reflected.” Autechre (AE) is an electronic music duo that is relevant to this disccusion in that they have gradually become sort of AI-infused, making music in tandem with generative Max/MSP software patches they’ve been refining since the 1990s, when laptops became viable music production tools. While I don’t know what AE’s source material is (raw waveforms? samples?) or how they work with it to create their sui generis music, there’s clearly a dialogue happening between the musicians and their code. But the reason I remembered “32a_reflected” has to do not with technologies or techniques but with beauty. In this track I can’t unhear a deep musicality infused into humming drones and an unusual form. Over seven minutes something seems alive: the music orchestrates compellingness, by which I mean it goes somewhere and makes you care about what’s going to happen next. This is an accomplishment of enchantment. AE’s tools are all digital, but somehow the touch of their craft–the ways of making, organizing, and listening–is everywhere tacticle in their sound.