A study on the continuous wavelet transform for the Vietnamese speech processing

An approach for the Vietnamese speech processing based on the continuous wavelet transform (CWT) is presented, Basically, the Vietnamese is a monosyllabic language using the roman characters. In particular, it has six tone-marks, which give it a singsong effect and different meanings for a word according to the tone-mark of the language. The spectrum of a word so is very complicated and the formants stretch through wide range in frequency domain. The proposed approach uses the technique of the local maxima of the transformed signal to extract the significantly detailed local as well as global information of the analysed speech. It is efficient for the processing and a suitable method due to the localisation feature and multiresolution of the CWT applied to the speech with a melody as Vietnamese. The simulation investigates a speech segment, which consists of sir tones of the word "ma" in Vietnamese speech, showing different characteristics of the signal in the transform domain.