Home

DSP STATE OF THE ART

##PITCH DETECTION

The Modified Cepstrum is implemented. A qualitative assessment on the pitch error is performed with the program OnsetAlgoTest.m which uses a set of piano notes from midi 23 to 107.
It can be seen that it perform bad after ~87.
Using this algorithm on the main.m lead in a bad estimation of pitch for high notes which are present in the test.

##ONSET
The test on a dual set of pitch windows were not successful. It consisted in using the a short window and a long window performing both a sliding pitch estimation. When the short estimation differs from the long one a note should occur.

Using an energy method the results seem to be accurate with few trial and error parameters selection. It performs well on them test with the program OnsetAlgoTest.m, however when the density of note become high, errors are occuring.
It also not a versatile method as parameters are set by trial and error.

##MAIN/Matlab If You run the main.m in the folder Matlab, you can see the results of the different algo.
It's quite fast, however the last part to replay is quite consuming (CPU) because it's a fast patch.
You can see that there is error in pitch, likely in the high frequency part because the original music the high pitched. There is also mistakes of note placements.

Data Markov extraction

Regeneration of music