Granular synthesis decomposes sound into particles — tiny fragments of 1 to 100 milliseconds — then reassembles them into textures that can be entirely unlike the source material. A human voice becomes a drone. A field recording becomes a rhythm.
What fascinates me is the parallel to how neural networks process information. Both granular synthesis and deep learning operate on the same principle: decompose complex signals into elementary units, transform them, recombine. The grain of a synthesizer is the token of an audio model.
In my live sets, I work primarily with processed vocals — my own voice fed through chains of granular processors, delays, and filters. There’s something compelling about being both the source material and the sculptor. The output is recognizably derived from something human but lives in a space that no human voice could occupy on its own.
This is what I think the best ambient music does: it finds the boundary between the organic and the synthetic and builds a home there.