Attached is the audio recording of my segment of Songs for Dynamical Systems, the workshop-plus-gig I recently mentioned, part of the Live Algorithms for Music network. The workshop itself was enormously stimulating, bringing up many interesting issues that only emerge when a software system is forced to improvise entirely autonomously with a human performer -- mostly the class of judgments which are entirely intuitive to a musician, and broadly without generalisable rules. Without any symbolic input, for example, how can an autonomous system gauge when to cease playing?

The gig itself, which was the culmination of a feverish few days' collaborative coding (using the P/F/Q = machine-listening/algorithmic-sequencing/machine-synthesis trio that constitutes the core Live Algorithm philosophy of modularity), showcased an impressively wide range of approaches and sonic palettes. It was also great to see collaborative performances pulled off with such panache, with OSC used for communication between participants over a wireless network.

My approach, audible in the recording above, was to develop a minimal responder that would roughly mimic Eddie's playing along a number of axes, each of which corresponds to a high-level musical feature (density of onsets, loudness, roughness/smoothness, frequency centroid, etc). Due to the noisiness of these measurements, and a certain degree of chaos inherent within the sequencing system, this is never remotely as linear as it sounds on paper! I also adopted a vaguely agent-based approach to positioning the sounds in stereo space, with a notional 9 agents responding autonomously to Eddie's playing.

Eddie himself was applying extended techniques to his tam-tam, with bowing and scraping supplemented by a motorised set of wire-beaters that he has affixed to it. It was an enthralling performance to watch, despite the nerves inherent in watching a bunch of piecemeal code attempting to play along with such a marvellous performer.

Thanks to Ollie Bown and Sam Britton for the recording.