OK as some of you may know @weefuzzy is now working on a new project we could describe as critical humanities of AI in music. He is tackling with Bob Sturm the ‘simple’ question of ‘musical genres’
I say ‘simple’ but it should be “(‘<simple>’)” where the famous “it depends™” is incredibly well explained and articulated in ways that are hard to ignore… and shows us ways to think about this in our modest practices.
Highly recommended. He is too modest to brag about it here, so I have to share!
Hey @weefuzzy : I want to share this video of a talk I watched at Strange Loop. Are you familiar with it?
This was prompted by your mentioning automated breakbeat analysis during the Q&A. This guy takes it further to talk about sample transformation. At one point he discusses the possibility of tracing specific variations of the same breakbeats over time as part of a musicological analysis.
I’m also reminded a bit of of Maria Perevedentseva’s talk at Timbre 2020 . How do you think about affect and affective discourse wihin your Mr → D → L model?
I’d not seen that talk but I have read his thesis, which is good (not least for having a good ethnographic component alongside the technical stuff). I’ll definitely watch – my dim memory is that what he was trying to do was really challenging for MIR methods: in some tunes the breaks are so chopped and mangled (and combined / layered) that it’s a tall order for a human listener (well, a non-aficionado) to pick it all apart. There’s a German language paper from the late 1990s where someone heriocally manually transcribes some breaks from Dead Dred and ShyFX.
I was very fortunate that Maria Perevedentseva was actually at that talk. Her work is great, and hopefully we’re going to talk more. The whole business of affect is kind of at the heart of it all, insofar as a driving question is why / how certain sounds get taken up and stuck to genres (or complexes of genres) the ways that they do, and the various ways that this signifies. Christabel Stirling’s work is great for thoughts on those lines as well.
I find dance music musicology really interesting and fertile for this (the ethnographic nuances). I was practically jumping out of my seat when I heard Maria’s talk. I felt almost… called out in a way because I’m so familiar with the weirdly specific insider music writing of the kind you read on Boomkat, the Wire and record shop releases lists.
Literally in the latest batch from Boomkat right now: deadly deadly deadly 70 minute session from cult UK club technician Carrier hailing his roots in mid-late ‘90s D&B with a ridiculously strong mixtape of pre-millennial, post-junglist rufige - 100% essential for fans of the golden era, the bruk out, Raime and Logos’ Reel Torque mixes, and anyone sick of being recommended Intelligent Liquid D&B to Chill and Study to.
It’s clearly important that it’s not once but 3 x DEADLY!!!
It’s a fair question at this point: was this written by a human? I think it must be, by the way it throws shade on casual YouTube listeners like that. Personally, I find it way more straightforward and reliable to glaze over (not even read) writing like this in order to discover music I would want to listen to over any existing recommendation algorithm (current listening: Finn, Dismal House mixtape).
I wonder what the application of this research could mean for various audiences and sectors of practice. What’s your vision @weefuzzy beyond better music recommendation algorithms and musicology papers? Would a socially conscious AI change the constitution of the audience and connect people in new and different ways? Would it change the design of commercial music production software also?
I sometimes wish that in my scene (of free jazz improvisors in Canada), there were more of a “dance music” sensibility to the way we communicated as musicians. Like maybe more specificity in the affective language that we used. I talked to Maria one time about an idea I had about an ethnography of rehearsal techniques (inspired by that one recording of Ornette directing his band that circulated a while ago). I’ve found in the past that improvisors are suspicious of talking too much about what we do even among ourselves.
Returning to dance music, I remember things like Ishkur’s Guide. Putting aside for now the issue of actually paying artists, I feel like a future “Spotify” could have geographic maps or a genealogical explorer type of UI. I don’t know if anyone would remember things like Microsoft’s Encarta Music Encyclopedia? Maybe nostaligia today but I feel like it’s maybe a more enriching way of discovering music than we have at present.
Some musings but it’s a topic that inspires me very much.
IIRC they just laid off the person who was responsible for doing a lot of the work in mapping out computational models that would explore the relationships between genres.
Oooh, biq questions! I think better recommendation and better music studies already have a lot to be said for them; e.g. one of the other projects on MusAI has been developing recommendation metrics specifically targeting a social goal of more / better communality (although it can’t do the job itself, obviously). Beyond that, my particular hobby horse is to erode the barriers between (technical) specialists and users and to consider the possibilities for this stuff beyond the profit motive that frames so much of the discourse around both music consumption and music software. For instance, how can different musical communities of practice have greater say and agency in the musical / social values their tools embody – again, not purely a technical issue, but the current dominant paradigm that sees the ‘best’ technologies as coincidentally aligned to the specific capacities and ambitions of massive data harvesting platforms is something academic research can and should push back against.
Lol – musicians may be among the worst people to turn to for finding affective language, not least because we can often communicate a whole lot in rehearsal via meaningful glances and grunts. It’s interesting what a massive role critics (however informally defined) play in keeping the language moving. I’ve certainly found with this jungle drum & bass work that a lot of the shifts in terminology as well as verbalisation of aesthetic priorities seems to arise not from producers / DJs but from journalists, promoters, labels…
There are three ecosystemic ways to approach the data-modeling of musical
genres: you can let artists self-identify, you can crowd-source categorization
from listeners, or you can moderate some combination of those inputs with human
expertise.
Two of those ways don’t work. Artists self-identify aspirationally, not
categorically.
From We Will Know Ourselves by Our Love; We Will Know You By What You Let Go furialog
This is fascinating and a lot to think about. I’m thinking about the different usages of genre terms and what Owen was saying about critics. I remember how, for example, “post-rock” (a term coined by music journalist Simon Reynolds) came to mean something completely different from what is familiar to most music fans. At the same time, I can’t think of anybody credible who would use it to describe their own music.
In the same spirit, any artist can be in as many different genres as apply. The
genres arenʼt even of the same sort: “tekno” is a very particular dancemusic
style, defined by tempo and historical circumstance; “wind ensemble” is a
configuration of performers; “Christian hip-hop” is philosophical distinction;
“Slovenian rock” a cultural and geographic one.
I’ve been thinking a lot about the notion of history as it relates to my practice since in jazz, tradition is considered very important (even if you’re somewhere on the margins, or intentionally running against the grain). There are various politicized notions of what the tradition even is, making “jazz history” politically contested. At the same time it’s widely understood even in the mainstream that the tradition is diasporic and geographically diffuse: that there can be resonance across distance and time.
I found the everynoise.com map pretty cool but also overwhelming. I did a bit of spot checking of “experimental” and adjacent terms and didn’t find the examples as compelling or illuminating as the dance music of regionally specific styles.
Like are historical narratives “latent spaces” in data science terms? Is it bias? Is this bias corrected by having a large enough pool of human experts validated the analysis? How are historians thinking about large corpus machine learning?
me too - at times surprising ‘rapprochements’ are inspiring, most time I find myself trying to find a sonic signature that would make sense - then I remember how Bob Sturm, to be cheeky, analysed a piece of mine in his algo during a talk I was present - I was impressed again, but it was a jump for my creativity more than any rigorous classification result - the human(s) in the loop got a spark to think…
And then, I am reading ‘Hungry Listening’ now, by Dylan Robinson, which gives me a good shock therapy on how listening to sounds as objects is such a Westerner’s approach… Not that I will change my practice of critical listening and composing with sounds, but understanding its deep biases and cultural anchors keeps me away from any sort of essentialisation and generalisation, and helps me deal with my issues with sampling…