TW374152B - Voice analysis system - Google Patents

Abstract

The voice analysis system 10 comprises a filterbank analyzer 18 capable of generating a data vector of continuous frequency from the voice signals of two speakers. From each data vector, the units 22A and 22B can compensate different forms of distortion in each speaker thereby generating a set of modified data vector. The computer 24 matches the modified data vector to the status parameter of an implied Markovian which can calculate the modified data vector exhibiting a maximum matching probability in each set, the matched model status parameters, the related distortion form, and the model type of a voice type or a noise type. The average value of the matched model status parameter is used to provide an estimate of the related data vector provided thereby. The estimate is compared with its related data vector, the difference thereof and the difference of other infinite pulse response filterbank 48A or 48B are averaged to provide a compensation of that type of distortion. A plurality of distortion forms of each speaker are provided by the averaged differential vector.