reg:exp

Sequencing Graphs

This week's idea: to build a new interactive music sequencer by adapting "Motion Graphs" (Kovar, Gleicher, Pighin, SIGGRAPH2002). Motion Graphs assemble believeable new walk paths from a limited set of motion capture input data by segmenting the data into small clips and creating a DG (directed graph) with clips as nodes and transitions wherever the beginning of one clip will seamlessly blend into the end of the previous clip as edges. Measures of similarity for mocap actions include joint angles and velocity data.

I now suggest taking the principle and substituting loops or other short music parts for the clips. By walking the graph, you chain a sequence of loops/bars into a linear time order. In a live performance situation, the artist can click on edges to control the sequence order; the system can also decide on a fitting or most likely next path automatically by comparing similarity measures between nodes to ensure continued playback if the user does not direct further playback.

Graph walks can also be specified in advance and recalled to pre-record sequencing before an actual performance. Finally, graphs could also be adapted for dynamic effects sequencing, where nodes control effects parameters for a specified amount of time. Cyclic graphs seem to be more appropriate than acyclic graphs for reuse of subparts and looping.