 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
FFT peak search-pitch detection experiment
Summary
| A Matlab / Octave example program that finds the peak in the spectrum of an audio audio file and gives the corresponding pitch. The simple approach works for basic waveforms, but is not a good pitch detection algorithm per se (see the comp.dsp thread). |
Matlab / Octave example program
Download note.wav
% *******************************************************
% FFT peak search example
% Markus Nentwig 2007
% This program is in the public domain
% Note: Peak search as shown works with basic waveforms
% such as sine, triangle, square, sawtooth etc, but is not a
% good general pitch detection algorithm
% *******************************************************
% read wave file
close all; clear all;
[y, fs, bits]=wavread('note.wav');
[n, nChan]=size(y);
if nChan > 1
error('this example requires a mono audio file');
end
% calculate the frequency corresponding to each FFT bin.
% this includes negative frequencies!
freqbase=fs*(mod(((0:n-1)+floor(n/2)), n)-floor(n/2))/n;
% apply window and get FFT. "Hamming" could be omitted
spectrum=fft(y .* hamming(n));
% get power (square of bins)
spectrum=spectrum .* conj(spectrum);
% pick out the audible range (for plotting)
bin1=min(find(freqbase > 16));
bin2=min(find(freqbase > 16000)-1);
spectrum=spectrum(bin1:bin2);
freqbase=freqbase(bin1:bin2);
% plot. Empty bin: use -150 dB, can't take log(0).
plot(freqbase, 10*log10(spectrum+1e-15)); xlabel('f/Hz'); ylabel('p/dB');
% peak search. Note: comparing floats by == is OK here.
% 'find' can return several results.
peakbin=find(spectrum==max(spectrum));
% look up the frequency corresponding to the bin
fPeak=freqbase(peakbin)
% For comparison: The fundamental pitch of note.wav.
% keyhole C => 9 semitones below 440 Hz "A"
fActual_Hz=440*power(2, -9/12)
The spectrum is shown in figure 1.
 |
© Markus Nentwig 2007-2008
The content of this page is provided without any warranty and may not be reproduced without permission.