Documentation
Guilleminot Christian
Translated by Loïc Corbel
What is PMPC ?
PMPC is based on the atomic decomposition of the signal. It allows the user to study, compare and classify non-stationary signal sequences. As voice is a non-stationary signal, PMPC is ideal to deal with voice corpora, but any kind of non-stationary signal sequence that can be named can be studied using this software chain.
PMPC treatment chain is complex because it requires high-end programs primarily designed for computering and applied mathematics. We started simplifying this chain by automating the most repetitive operations.
As of now the automation does not concern the choice of sequences before being processed with Complearn, as this choice belongs to the researcher. However, the user can use simple Shell scripts for this part as well.
Use in phonetics
First, Praat software is used to tag sound sequences. PMPC adds buttons to the Praat interface, which are used to start the decomposition of all tagged sequences and put them into a given folder that can be automatically created.
At this point each sequence appears as a file containing numerical values of every atom.
From this step of the process, atom files can be classified, selected and processed with Complearn.
Complearn produces a distance matrix of sound sequences that can be visualized as an image of a .pdf graphic. The method consists of a similarity calculation based on Kolmogorov complexity.
Other uses
Every wave-like sequence can be processed using PMPC, wether it is from a natural phenomenon or transformed.
Reliability of PMPC
MPTK and Complearn softwares have been successfully used in many analyses.
MPTK is used in shape recognition, signal compression and signal sources partition.
Complearn is used in molecular biology, linguistics, zoology, shape partition.
Why should you expect better results compared to other methods?
* MPTK decomposes the signal into atoms, which gives a better controlled description of it compared to Fourier's methods. ;
* The atoms are sort of music notes : a group of atoms can be considered as the signal's music score. ;
* Music score comparison, notably using Complearn, gave superior results to any other method to identify music genres and composers. ;
* The only classifications that gave experimentally reliable results were based on written scores or MIDI files (MIDI files are written scores translated in computer language).).
Other uses : PMPC has been designed to solve phonetics problems, but can be applied to every digitized waves such as :
- Bird cries and songs, animal cries ;
- electroencephalography ;
- Noise detection, surveillance and monitoring ;
- Materials tests...
Classification according to J K M Y with the /u/ vowel and /s, sʕ, d, dʕ/ consonants
