Sound Examples.
A synthetic 4-note arpeggio is analyzed with a single FFT. Spectral segments are identified around amplitude peaks and grouped by group delay similarity. The temporal order of the groups is then reordered, so the notes play in a different order while preserving their original amplitude envelopes.
Synthetic arpeggio (original):
Synthetic arpeggio (reordered):
The same transformation applied to a recording of a glockenspiel playing a 4-note arpeggio. The inharmonic partials of the glockenspiel are correctly identified as belonging to different notes and reordered in time.
Glockenspiel arpeggio (original):
Glockenspiel arpeggio (reordered):
A single harpsichord note is analyzed. Each spectral segment is displaced in time by an amount proportional to its peak amplitude. Higher amplitude partials are displaced further, causing the spectral content to unfold over a longer duration. The per-bin example applies the same displacement to every frequency bin independently, without segmentation, producing the smeared texture characteristic of incoherent giant FFT manipulation.
Harpsichord note (original):
Harpsichord note (per-bin displacement, no segmentation):
Harpsichord note (per-segment displacement):
Sinusoidal modulation of group delay values across frequency creates discrete temporal copies of each spectral segment. Different segments are modulated at different rates, with lower frequency segments repeating slower and higher frequency segments repeating faster. The output is extended to 20 times the original duration.
Harpsichord note (displaced and modulated):