A robust speech endpoint detection algorithm based on wavelet packet and energy entropy

The robust speech endpoint detection is important in many areas of speech processing. In this work, a robust speech endpoint detection algorithm is proposed based on wavelet packet and energy entropy. The proposed method is tested using speech utterances under four different background noise: White, Pink, Babble and Volvo with different SNR (-5db, 0db, 10db, 20db). The results are compared with traditional short-time energy plus short-time zero crossing rate method. The experiment results indicate that the presented algorithm achieve higher endpoint detection accuracies than traditional method. The correct detection rates reach up to 97.5% for noise-free speech, and range from 80.72% to 93.87% according to different types of background noise.