The fftExtract software is a simple wrapper around fftw for extracting feature vectors from sound files. It is primarily intended for use with audioDB.


fftExtract is a command-line executable only; its syntax is

fftExtract [options] inFile outFile

where the options are:

feature options (mutually exclusive):
-m M : MFCC (Mel-Frequency Cepstral Coefficients)
-q B : constant-Q spectrum with B bands per octave (0=linear freq.)
-c C : chromagram folded into C bins
-P   : overall log power

FFT options:
-l L : low edge of constant-Q transform (63.5444Hz = low C - 50cents)
-i I : hi edge of constant-Q transform (8kHz)
-n N : length of FFT
-w W : length of window within FFT
-h H : hop size

Extraction options:
-C c : channel (0=left, 1=right, 2=mix)
-p file : fftw plan file (generates new if does not exist)
-b file : beat timing file (secs)

Debug options:
-v V : verbosity [0 - 10]


The source code is available under the terms of the GNU General Public Licence.



The fftExtract software is actively maintained; a gitweb view of trunk development is available here; please report any bugs to the audioDB mailing list [Mailman info page].


The development of fftExtract was supported by the OMRAS2 project, funded by EPSRC (grant number: EP/E02274X/1).