Voice signal dynamic feature extraction method based on MUSIC and modulation spectrum filter

Abstract

A voice signal dynamic feature extraction method based on MUSIC (multiple signal classification) and modulation spectrum filter mainly comprises voice signal preprocessing, MUSIC spectrum estimating, modulation spectrum filter, spectrum energy modulating and calculating, log energy calculating and discrete cosine transformation (DCT). The modulation spectrum filter can not only fully reflect dynamic features among voices but has low sensitivity of voice environment owing to the modulation spectrum filter has a time-frequency clustering, so that according to different reflections of interference signals and voice signals on the modulation information, the modulation spectrum filter uses spectrum estimating technology based on multiple signal classification, modulates the spectrum filter of obtained MUSIC spectrum and extracts cepstrum coefficients of the MUSIC spectrum as feature parameters. Compared with the prior art, the voice signal dynamic feature extraction method based on MUSIC and modulation spectrum filter has good robustness and not only significantly improves the recognition rate of a voice recognition system but has played a good foreshadowing role for follow-up studies of voice signals.

Description

technical field [0001] The invention relates to a method for extracting dynamic feature parameters of Chinese speech signals, in particular to a method for extracting dynamic feature parameters of speech signals based on MUSIC and modulation spectrum filtering. Background technique [0002] The most basic and most important development link of speech recognition is the extraction of speech signal characteristic parameters. As early as the 1940s, R.K. Potter and others proposed the concept of "Visible Speech", pointing out that spectrograms have a strong ability to describe speech signals, and tried to use spectral information for speech recognition, which formed the earliest phonetic features. In the 1950s, people found that in order to recognize the speech signal, some parameters that can reflect the speech characteristics must be extracted from the speech waveform, which can not only reduce the number of templates, the amount of calculation and storage, but also filter ou...

AI Technical Summary

Problems solved by technology

However, once these parameters are applied to noisy environments, their performance will drop sharply

[0003] Moreover, the characteristic parameters mentioned above all reflect the static characteristics of the speech. The dynamic characteristics of the speech signal refer to the characteristic parameters extracted from several consecutive frames of speech. Acceleration parameters cannot fully mine the dynamic information, so they cannot reflect the dynamic characteristics of the speech signal well

Images