Notes from Underground

Notes from Underground, a sonification of London tube station data, was, at first glance, a simple port across to Web Audio. The piece uses live arrival information from the TfL API. Rather than translating those data into pitches or melodies, each train is assigned a unique “chord” of sounds drawn from field recordings made on the Underground itself. Three digits from the train’s ID number select samples from banks of voices, train sounds and station ambience. These samples are then granulated, with playback position tied to the train’s estimated arrival time, producing an irregular, accelerating time-stretch (influenced by Carl Stone’s work). As trains draw closer to the station, their sounds gradually emerge from the texture, creating a continuously evolving sonic portrait of the network.

But there was a snag in just bringing it over to the web as-was: I presented it in my PhD as a work without a home. There were a couple of implementations of the original piece. One picked a fixed (reliably busy) station, and another allowed the user to select a station. Both presented more like demonstrations than works in their own right, and neither seemed an especially satisfying direction for the webpage. They offered either pure randomness or a kind of unsatisfying point-and-click interactivity that, whilst taking advantage of the affordances of the web, nudges it toward playable online instruments or even game territory. Both perfectly valid directions, but not the direction of this work.

When making the original piece I had a vague idea that, in an installation context, many stations could be sonified at once (on different machines), for example arranging stations from a specific line spatially across a gallery space. I would have loved to try that. I even imagined a version built around a giant tube map poster, with each station circle replaced by a headphone port. Gallery visitors could literally plug in and listen to a station. The technical challenge of that is considerable - a supercomputer playing every station simultaneously to a monstrous multichannel output, or some sort of clever switching multiplexer that tells the system what’s plugged in and what to play. But for the web?

The most obvious implementation was “random station”: open the webpage and a random station is chosen by the code and sonified. It produced some uninspiring results. Interchanges and busy stations can have twelve or more trains to sonify at any one time (actually forcing voices to be limited due to the number of audio objects required). In contrast outlying stations on single lines can have one train every five or ten minutes. Now, a “boring” result has an experimental, Cagean chance-operation purity to it, and my response could be to simply accept the results, but that doesn’t fit with how my sonification work has evolved. I think there’s a compromise to be made between that practice of “just accepting the results” and “staying true to the data”: not applying excessive transformations or transpositions to deliberately engineer an outcome, but instead selecting from the data in such a way that the piece remains active and interesting.

I experimented with a couple of iterations (frustrating my AI coding assistant considerably and producing some confusing results - I frequently found trains stuck, presumably waiting for a platform, just seconds away from the station, which plays havoc with the time-stretching device). After much tinkering I arrived at a system which selects eight stations to begin with, sonifies the nearest train, and then waits for the next reading of the API to select a new station. It risks being read as a gimmick, but in designing this version to effectively “rove” around the network I ended up with a unexpected new reading of the old work. This trial-and-error-derived version injected some breathing space into the piece. It gave it a curious flow that I liked. There are moments of great activity as several stations crescendo at once, followed by strange calm as the next trains are far in the distance. Listening to it over a long spell, it sustains itself remarkably well, with chance results sounding spookily composed.

I found myself spending more time than strictly sensible on the visualisation, but I’ve come to see these as a necessary evil of the website. When trialling Dawn Chorus years ago, one comment was that it needed some sort of visual to persuade visitors that something was going on. Thus I’ve added something visual to each one of these resurrections. I avoided doing so when submitting my PhD pieces, relying on a clean, technical Pure Data front end, feeling that a flashy visualisation would take attention away from the sonic aspect. Whilst I’m confident of my sound-art credentials, I am not a graphic artist. Happily the tube map (although copyrighted by TfL) is a design classic and surely ripe for exploitation here - being itself a sort of schematic data transformation. With some hacking I reproduced the coordinates of the tube map, shorn of lines and labels, and turned it into a sort of moving constellation or light board. Currently stations light up. Neat. Minimal. Elegant, even. And a real pig to program, even with AI assistance!

The original version of Notes from Underground is documented in my PhD thesis.