Driving SoundSpotter live in Max/MSP and PureData

SoundSpotter ships as an external for both Max/MSP and PureData, built on Thomas Grill's FLEXT layer. On stage, the trick is less about the object and more about what you feed it and how you steer.

Load a corpus, pick a target

Start simple: one corpus buffer, one live input as the target. Keep the corpus focused. A tight, characterful corpus (one instrument, one session) usually sounds better than a giant grab-bag, because every match lands inside a coherent sound world.

Steering controls

The interactive controls are the instrument. A few that repay practice:

Latency

Keep your block size sane and the feature window modest; SoundSpotter is highly optimized, but the search still has to finish inside a buffer. If you hear it stutter, shrink the corpus before you blame the object. For build help, see the source-build notes.

← Acoustic lexemes and the TimbreScoresFrom SoundSpotter to audioDB: scaling similarity search →