5.5 Examples and code

These are the files we used to implement our Pitch Correction&Detection system, along with some examples of our results.

Matlab code

When using these programs, type help<name>into MATLAB to determine the format of the inputs and outputs. Vectors are always columns, and sounds are always monaural. Sounds should also be normalized. In general, vectors of pitches refer to windows, and the pitches themselves are in Hertz. Advisory: choose window size of 4000 samples and jump size of 1000, assuming a 44.1 KHz sample.

Pitch detection

fastauto.m

findautomin.m

Detect using the frequency-domain HPS algorithm:

hps.m

Determine the pitch to which the program should correct:

Pitch determination

Pitch correction

tdrev.m

Correct pitch using the PSOLA time domain algorithm:

psolarev.m

Correct pitch using the Modified Phase Vocoder:

modpvshift.m

Examples

These examples were all generated using the FAST-Auto pitch detector with threshold around 0.05, a 4000 point window, and a 1000 point jump size, using the PSOLA correction algorithm.

Flat correction

Original Corrected

Scale

Original Corrected

Song adjustment

Here, listen to how the corrector operates on this song. The singer was not out of tune to begin with, but there were points at which he deviated from a perfect note and the corrector adjusted his pitch. Corrected

In this last example, we use the correction system in the wrong direction: for distortion! We took the original clip above and attempted to make the singer monotone by demanding the same key for the entire clip. Listen to the results: Adjusted