Efficient simultaneous multi-scale computation of FFTs
An efficient algorithm for simultaneously computing FFTs at multiple window sizes is introduced. It requires 42% fewer complex multiplies for an 8192-sample window. It is motivated by STFT visualization of non-speech audio events, which can vary widely in scale and resolution of both time and frequency. A reference implementation in C agrees with MATLAB's implementation, for numerous test inputs.
This software was published in 2009 as a technical report of the NSF program Foundations of Data and Visual Analytics.
To build and test
sudo apt install make g++
make test
To regenerate the report's figures
sudo apt install imagemagick ruby
sudo apt install gnuplot-x11 (or -nox or -qt)
sudo gem install gnuplot
make figures
To cite
Dave Cohen, Camille Goudeseune, and Mark Hasegawa-Johnson. 2009.
"Efficient simultaneous multi-scale computation of FFTs."
Technical report GT-FODAVA-09-01.