Implementation of concatenative sound synthesis based on matching four features from a control input to the database generated from a source input. The control determines which frames of the source get played back, and you can change the weighting of features to refine your matching criteria (ie, make rms the most heavily weighted to have amplitude as the most important criteria). You can also modulate the match length, being the amount of source played back for a match, the feature weightings, and freeze the database collection to work with a collected sample from the source.
Refs- see the publications of Schwarz, Sturm, Casey amongst others.
control |
audio rate input, acts as control |
source |
audio rate input, source for cross-synthesis |
storesize |
size of source store sample buffer in seconds |
seektime |
Time in seconds into the past to start searching for matches |
seekdur |
Time in seconds from seektime towards the present to test matches |
matchlength |
Match length in seconds (this will be rounded to the nearest FFT frame) |
freezestore |
Stop collecting novel source input, keep store (database) fixed |
zcr |
Weight for zero crossing rate feature |
lms |
Weight for log mean square amplitude feature |
sc |
Weight for spectral centroid feature |
st |
Weight for spectral tilt feature |