The self-adaptive voice activity detection algorithm based on time-frequency parameters

Abstract

In order to solve the inferior performance and sad self-adaptive of the traditional voice activity detection algorithm in an environment with low Signal to Noise Ratio (SNR), a new self-adaptive voice activity detection algorithm based on time-frequency (TF) parameters is put forward. After introducing the time-domain log-energy and improved mel-scale log-energy, the new TF parameters are acquired by coalescing them, which make it possible for distinguishing speech from noise effectively. Then, the TF parameters are updated to predicate endpoint through the threshold test. Finally, simulation experiments show that the algorithm can improve significantly the performance of automatic speech recognition (ASR) system and robustness. When the SNR is 0dB, the error rate of the algorithm is about 15%.