Photographic Punchlines

I’ve moved slowly, and nowhere but inside my apartment, for an entire day. It is late in the afternoon, and I am at my desk meditatively typing away, answering emails and fixing small bugs and watching the cats lounge in the long sun angling through the open windows. I’m adrift in casual relaxation as the summer blows its breezes around. The phone goes off. It is my landlord. “Hey, move your car the contractor can’t get into the trailer he just put in the driveway.” I expect there to be three or four cars in the driveway, and that I’m blocking folks in.

 

The next day, also exceptional in its perfect summerness, I am ambling through a park – never further from a hurry. A burbling finger of the Waloomsac river fills the air with that unintrusive aural incense, as do the clouds of rustling oaks all around. Walking the path from one end of the park to the other, my blissful disregard is slowly overwhelmed by concern. Growing louder and louder are the shrill cries of children, hundreds of them. They sound terrified, as if a gunman has stormed their school. The trees obscure any hint of the horror. I’m walking closer, quickening my pace, the screams swelling in volume, and know that regardless of the situation I will be utterly unprepared for what I see.

rpm + Bandcamp = meta-band

Since joining Bandcamp, I’ve been eager to find the time and functional knowledge to start making visual and aural representations of different sets of Bandcamp data. After dithering, toying, considering, and learning, I am about to finish up a piece of music (if you’d like to call it that) using three months of international sales data. The currency, value, and time of nearly a hundred thousand sales are used to control parameters in the audio domain. I like to think that if Tufte were a musician, he would make tunes like this.

Here are some exploratory thoughts at the tail end of making:

The sales are so frequent that it is easy to consider them on a larger scale: not just granular moments, but as the continual state and process of sales. Musically this translates into a hum, a continuous buzz. Give the things known about individual sales, how might they be sonified in terms of that hum of which they are a part?

Each small unit of sale, the atomic element of the dataset, maps onto the atomic, discrete unit of audible sound: the grain. The frequency and amplitude of that grain are tangled with the properties of the sale in appropriate-ish ways. The amplitude is coupled with the relative value of the sale (its cost), and the frequency is related to the relative frequency of the currency type in the dataset.

There is plenty of room for fine craftsmanship and artfulness in the way these properties are explicitly mapped. Rather than assign tuned pitches -notes in a scale- to the currencies (as was my initial choice) I’ve decided instead to merely set the upper and lower boundary for frequency, and map the relative frequency of the currencies as a continuous function over the audible spectrum. This adds even more information about the sales to the audio representation, without having to modify the formal constraints of the music. I may map the frequency of currency-type logarithmically over pitch to be easier on the conditioning of the listener’s ear (that is, the typical pitch distribution we hear out in the world, and how our vestibular system developed around that). The idea here is not to demand or necessarily encourage a different kind of musical sensation and/or listening, but rather to make the nature of the continuous state of sales perceivable in a domain of incredibly fine resolution. And to pull that off in such a way that listening is even pleasurable.

Little is fixed in the mapping; the way quantified sale properties correlate to corresponding auditory attributes is dynamic. The relative frequencies of the currencies & the range of least to most expensive sale are different in any given window of time, and so change. Is the scale of the mapping in the piece recalculated on a day-to-day basis? Month to month? Fully continuous? How would the latter be executed? Rather than modify the scales in discrete temporal chunks in one recording (ie, a different range for each month in a recording representing three months of sales) I find it appealing to use a healthy sample of 3-5 months of data, with relative frequency calculated over the entire dataset. I imagine this will likely accentuate unusual sales and, as one and the same, help avoid the pitfall of monotony.