Some Statistical Study on the Short-Time Spectrum of Unvoiced Pronunciation Signal

For the signal of human unvoiced pronunciation, the amplitude value in short-time spectrum is studied statistically for each frequency component. A novel statistical feature is discovered for the discrete short-time amplitude spectrum, which reveals the statistical correlation between any two frequency components. Based on this statistical feature revealed, a new synthesis method for unvoiced pronunciation signal is implemented, and the synthesis results indicate the validity and effectiveness of the proposed method.