Single Channel Speech Boundary Detection Algorithm Based on Principal Component Analysis

In many speech processing fields such as speech enhancement, speech coding and speech compression, voice activity detection is of great significance, so it is often necessary to conduct speech boundary detection (SBD) in speech processing. The common characteristics of voiced signals are zero energy product (ZEP), linear prediction analysis (LPC), etc. Although these two typical features are widely used for noisy scenes, the accuracy is not enough. In this paper, each speech frame's mel-frequency cepstral coefficients (MFCC) are analyzed by principal component analysis (PCA); then, PCA is used as the feature vector of the current speech. The experimental results show that the algorithm accuracy of ZEP, LPC, PCA is relatively similar under the condition of a high signal-to-noise ratio (SNR). When the speech signal is disturbed by high noise, the accuracy of ZEP decreases, and the PCA algorithm can be used as one of SBD algorithms.