The application of wavelets vector quantization of Polish speech

The paper presents a concept of vector quantization of words uttered in the Polish language. Columns or rows of the matrix obtained as a result of time-frequency analysis of chosen words are vectors used for the further analysis. As a tool in the process of vector quantization was used the Wavelet Packet Transform (WPT) in which the signal decomposition scale is similar to the mel frequency scale (see method - Mel Frequency Cepstral Coefficients – MFCC). Such analysis allowed us to choose the best useful properties for the word recognition. Both column (in time) and row (in frequency) analysis are formulated in the form of computer procedures and compared. We hope such studies will be a starting point for further work on the system Automatic Speech Recognition (ASR).