Dynamic feature extraction for speech signal based on formant curve and MUSIC

摘要

In order to improve the robustness of speech recognition in noise environmental conditions, this paper proposed a new dynamic feature extraction method based on formant curve and Multiple Signal Classification (MUSIC) spectrum. It uses Hilbert-Huang transform to estimate speech signal formant frequency characteristics, and then gets the first formant curve by combining the first formant frequency characteristics of each frame from the first frame to the last frame, and so forth, gets the second formant curve, the third formant curve and the fourth formant curve. And then calculates the MUSIC spectrum and the energy spectrum for each formant curve, takes logarithm transform and discrete cosine transform. Compared with the method of MFCC, the proposed dynamic feature of speech signal has the time correlation, reveals the close correlation between the speech signal frames, improves the performance of speech recognition.