SWIPE' pitch estimator, v. 1.0

Based on Camacho, Arturo. A sawtooth waveform inspired pitch estimator for
speech and music. Doctoral dissertation, University of Florida. 2007.

Implemented in C by Kyle Gorman <kgorman@ling.upenn.edu>

HOW TO CITE:

Please cite this dissertation, and if possible include a URL to this program
source, the most-recent version of which will always be at:

    http://ling.upenn.edu/~kgorman/c/swipe/

HOW TO INSTALL:

Read the included INSTALL file

AUDIO FILE FORMATS:

All mono-channel audio recognized by libsndfile is accepted. For a list, see:

    http://www.mega-nerd.com/libsndfile/#Features

A1. SYNPOSIS:

swipe -i FILE [-b FLIST] [-o FILE] [-r MIN:MAX] [-s ST] [-t DT] [-mnhv]

-i FILE     input file (must be specified
-o FILE     output file (default: standard output)
-b FLIST    batch mode [FLIST is a file containing one whitespace-delimited
            "INPUT OUTPUT" pair per line] (not compatible with -i/-o)

-r MIN:MAX  pitch range in Hertz (default: 100:600)
-s THRSHLD  strength threshold [0 <= x <= 1] (default: 0.3)
-t SECONDS  timestep in seconds [must be < sf/2] (default: 0.001)

-m          Output Mel pitch [= 1127.01048 ln(1 + hz / 700)] (default: no)
-n          Don't output voiceless frames (default: no)
-h          Display this message and output info, then quit
-v          Display version info, then quit

A2. OUTPUT:

pitch_0 time_0
pitch_1 time_1
... ...
pitch_N time_N

A3. A Python wrapper, swipe.py, is also included:

    class swipe --> __getitem__, __iter__, __str__, bounds(), slice(), regress()

It requires the stats.py and pstat.py packages, available at the following URLs:

    http://www.nmr.mgh.harvard.edu/Neural_Systems_Group/gary/python/stats.py
    http://www.nmr.mgh.harvard.edu/Neural_Systems_Group/gary/python/pstat.py

By default, the binary for swipe is "swipe": this will only work if the binary 
is in your $PATH. 

Example code:

<code>
#!/usr/bin/env python
from swipe import swipe
(start, stop) = (.2., .6) # a region of interest
pitch = swipe('test.wav', 75, 500) # read in pitch track via swipe
pAt1 = swipe[.6] # get the pitch nearest to 600 ms
pitch.slice(start, stop) # inline, chop it down to the interval [.5, .7]
print pitch.bounds() # prints the closest boundaries
(slope, intercept, r2) = pitch.regress() # lin. regression on sliced interval
</code>

A3. I also included the original MATLAB code from Camacho as swipe.m

A4. Questions, comments, bugs? Email kgorman@ling.upenn.edu