First, I don’t think I knew that Max Mathews was still alive.

The pioneer of computer music demonstrates his Radio Baton instrument in this half-hour video. The instrument (or controller, if you want to split hairs) consists of two batons with radio transmitters at the tips, which you wave over a panel that can sense their locations in three dimensions. That gives you six variables you can use to control whatever you want.

Mathews says:

The model that I’m using for this control is not the model of a violinist or a pianist playing a score. Rather, it’s the model of an orchestra conductor conducting music. The orchestra conductor doesn’t have to play all of the notes in a piece, or any of the notes, but he can control the expression of the piece and get the musician to follow his batons.

It’s an interesting model of technological music production, “playing” preloaded scores. For those of you who have been following along, it may be interesting to note that the “expressive” parts of music Mathews describes in the video (tempo and volume) are exactly what you could control on the old Pianola. Is Mathews “playing” the music like a discman or like an instrumentalist? Substituting “conducting” for “playing” is an interesting way to play on pre-existing arrangements of musical labor and validate his work. Now I’m curious to see whether the idea of “conducting” music was explicitly use by proponents of the player piano in the first part of the 20th century…

There is much more information about the idea of a “conductor program” at Stephen Malinowski’s page on it, where you can see some videos and download an application that lets you tap through MIDI files. (A similar idea is at play in Smule’s Magic Piano iPad app.)

(via @CompMusicBlog)

This is really neat: a history of PC audio, showing the audio cards and playing the corresponding version of the theme from Monkey Island.

I am somehow nostalgic for the time when you had to compose the music specifically for the hardware. But only a little: can someone suggest a way to compose specifically for contemporary computer audio hardware? There must be some way to get past its “universal” ability to reproduce sound and find its specific weirdnesses. (This might also help augment this history that stops in the mid-90s once “CD quality audio” arrives, as if the history has been nothing but an increase in fidelity and CD quality represents a reasonable endpoint.)

(via Retro Thing)

This is an interesting idea (although the video does seem to drag after a while): use some sort of magic microphones to capture the sounds (and I think magnetic interference) of a computer in operation.

This applet analyzes MIDI files for repeated forms, and then outputs an image like the one above, linking repeated parts. The image above is a mazurka by Chopin, with a “complex, nested structure.”

There is an image gallery worth checking out just for the extremes of repetition and non-repetition: Glass and Schoenberg are in there.

(via the music of sound)

In C In Live

I recently spent a little time trying to make a version of Terry Riley’s In C using Ableton Live.

The original piece, if you’re not familiar with it, is a landmark of aleatory minimalism—a score with 53 short fragments that are repeated by the players in an ensemble of varying size. Each fragment is repeated until a player decides to move on to the next. The collective result of these individual decisions is a vibrant pulsing mass of sound, with emergent melodies and rhythms coming from the phasing and combination of the different fragments. You can see a bit of it here:

Riley specifies that the players try to stay within about three fragments of each other, and to come into unison periodically, for dramatic effect.

To approximate this in Live, I used “follow actions.” Follow actions let you specify what happens when a clip (in this case one of the 53 fragments) finishes. You can choose two behaviors and assign them probabilities. So my first setup: when a clip finished, there was a 10:1 chance it would repeat. Otherwise, it would advance to the next clip.

The main distinction between a track in Live and a human player here is that the track has no knowledge of what the other tracks are playing. So, Riley’s admonition that “It is important to think of patterns periodically so that when you are resting you are conscious of the larger periodic composite accents that are sounding” is a little tough.

1. Since the players have no knowledge of each other in Live, they can only advance based on probabilities; this means that the clips tend to drift apart over the course of playing (I noticed at one point clips 15 fragments apart playing at the same time!)
2. It appears that part of the reason to stick together in In C is that fragments that are far apart can actually be dissonant with each other, while local groups are generally well-matched. So, this drifting is actually problematic from a basic listening perspective as well as a “doing what the score says” perspective.
3. Also, using the same probability for all clips means that shorter clips will advance faster than longer clips, encouraging this spreading.

I tried messing around with varying the probabilities throughout the piece, basically to make little speed bumps, where the probability of advancing past fragment 11 is low, so the clips catch up with each other. The problem with this is that it then holds on to a few clips for too long, spreading everyone out again.

I’ve uploaded the Live file for people with Live to check out. It has 35 voices as recommended in the score, all using piano variously panned for easiness; you’ll want to freeze the tracks to play it unless you have a monster processor. The quick fix to the drifting problem seems to be to just go around triggering clips to keep everything in sync. Try experimenting with different follow action probabilities!

I haven’t uploaded an audio file because I haven’t gotten it to a place where it sounds good leaving to computer to its own devices. This is really a job for Max/MSP, which would let me make tracks that are aware of the positions of other tracks, but that’s a project for another procrastinatory night!

[Flash 9 is required to listen to audio.]

Amerie’s “One Thing” featuring Michael Jackson’s vocals (DJ API)

Made by analyzing the vocal track from the original song, and then pulling tiny bits from a collection of MJ acapellas using an algorithm that finds the closest “match.”

related to the post below

(via Music Machinery)

Demoing the Computational Remix

The video above was produced using the Echo Nest Remix API to play every beat in “Boom Boom Pow” backwards. I have to say first, I am a big fan of the results. (And no, not just because I may have an unhealthy obsession with Fergie.) I think that for a demonstration of the Echo Nest’s beat-shuffling capabilities, a song so obsessed with the technological (the cliché technological in particular) and its own beat is a great choice. You can see some other basic manipulations over at Paul Lamere’s blog where he has cataloged a few.

Reviewing the musical results of computational processes like these is tricky business: are you evaluating the work of a computer? a person? just “the work” on its own? In this case, the work is not presented as a “work” per se, but just as a demonstration—look how easy it is to move beats around with relatively few lines of Python code, try making your own artistic productions, and so on. This is the special sneakiness of the demo: it is always “just a demonstration,” apologizing for simplicity with a technological story. “Pow Boom Boom” is not a song, it is an example of a bit of code, so my real aesthetic pleasure at listening to it is not the enjoyment of a song, but rather the enjoyment of a different kind of artifact.

So we have a question of packaging. The “demo” wrapper is armor against aesthetic critique, although I don’t think this computational remix needs it. Other uses of the Echo Nest Remix API so far (like MoreCowbell.dj or DonkDJ) are more aesthetically questionable, to say the least, converting songs from one genre to another computationally or running cowbell and Christopher Walken all over a song. Of course, these examples rely on another type of wrapper, “comedy.” Putting a Sigur Rós song into DonkDJ and turning it into bastardized Northern British lower-middle class dance music is surely a pleasure that would have been unimaginable for the creators of the ARPANET. I am reminded of using early consumer-level speech-to-text software in the 90s and laughing as dog tooth buffalo tomorrow.

While I was writing this post, a new example of the Echo Nest software’s application came out: a really promising set of tracks from the Echo Nest’s own DJ API (a.k.a. Ben Lacker). These tracks, presented as part of an actual DJ set, are more musically complicated than the beat-shuffling demos. By “musically complicated” I mean both that they move more into the territory of “real songs” and that they are more intricately structured remixes. These tracks are in between a demonstration and a full-on “song.” To a certain extent, any of this early work has the protection of “demo” on it (which is why I said “promising” before, rather than “great”), but these pieces are clearly trying to make early statements on what this kind of art can be like. There is promise here, but much of the appeal still lies in the novelty alone (not that I have a problem with novelty for novelty’s sake).

The most exciting piece of the group for me is a remix of Amerie’s “One Thing” in which Amerie’s vocals are replaced by snippets of Michael Jackson’s voice, sampled from a large set of acapella tracks. The Echo Nest Analyze API chewed through the Amerie vocals and used the analysis data to substitute the most similar-sounding bits of MJ, making a totally nonsensical lyrical line that vaguely follows the pitch and timbre of the original. This is like seeing a computer try to speak human language, and it is simultaneously terrifying and awesome. I think aesthetically, it is a great use of the tool (better than the version that retains Amerie’s vocals, replacing the scattering hits and drums that define the song’s aesthetic with samples from a gamelan).

I have some more things to say about what I think “most similar-sounding” means in the context of contemporary recordings and the API, but those will just have to wait for the next post.