wu_thumb.png

The Wu pitch tracking system

This is a Matlab implementation of the Wu pitch tracking algorithm [1].

This Matlab implementation was initially started by Keansub Lee and updated by Byung Suk Lee and Dan Ellis all at LabROSA.

You can download a zip file containing wu_main.m and the other files used in this demo from wu_v1.0.zip.

The input audio file can be in various formats, e.g. wav or sph. Note that the path of audioread, 'arpath', must be edited to point to the correct location that is the path of installed audioread. For example, our local (default) audioread path is '../audioread' .

Contents

----------------------------------------

OBJECTIVE:

         Perform pitch tracking on the input audio file using
         the Wu algorithm.

INPUT:

         filename - the name of the input audio file
         dispflag - flag for plotting the pitch detection (default: 1)
         prct - percentile (frame-wise) for the unvoiced audio
                (default: 61%)

OUTPUT:

          pitch_track - the output f0 in Hz.
          logprob - log probability of period (lags).

REFERENCE:

          [1] Wu, Wang, and Brown, TSAP, v11, n3, 2003
----------------------------------------

USAGE:

This package includes an example audio file, 'v3.wav', to try:

    [pitch_track,logprob] = wu_main('v3.wav');
Using Slaney-Patterson filterbank, frq=100..800, bpo=16
Autocorrelogram (maxlags:400, nchs:48)
(1/1) Start: 1 End: 23101
islice: 1 len_tmpac: 142 lac: 142
Elapsed time is 4.449827 seconds.
HMM pitch-tracking for Wu algorithm
Elapsed time is 0.059459 seconds.

This will generate a plot that shows (1) spectrogram of the input audio overlaied with the output f0 (Hz), and (2) the summary autocorrelation overlaied with the output period (lags).

To adjust (unvoicing) prct(entile) parameter to favor unvoicing:

    [pitch_track,logprob] = wu_main('v3.wav',1,87);
Using Slaney-Patterson filterbank, frq=100..800, bpo=16
Autocorrelogram (maxlags:400, nchs:48)
(1/1) Start: 1 End: 23101
islice: 1 len_tmpac: 142 lac: 142
Elapsed time is 4.512434 seconds.
HMM pitch-tracking for Wu algorithm
Elapsed time is 0.056060 seconds.

----------------------------------------

CONTACT:

          Byung Suk Lee, bsl@ee.columbia.edu
          Dan Ellis, dpwe@ee.columbia.edu
          LabROSA, Columbia University, 2012-09-13

Valid HTML 4.0! Last updated: $Date: 2012/09/16 22:18:57 $
Byung Suk Lee <bsl@ee.columbia.edu>