Study on the MFCC similarity-based voice activity detection algorithm

The accuracy of voice activity detection (VAD) is one of the most important factors which influence the capability of the speech recognition system, how to detect the endpoint precisely in noise environment is still a difficult task. In this paper, we proposed a new VAD method based on Mel-frequency cepstral coefficients (MFCC) similarity. We first extracts the MFCC of a voice signal for each frame, followed by calculating the MFCC Euclidean distance and MFCC correlation coefficient of the test frame and the background noise, Finally, give the experimental results. The results show that at low SNR circumstance, MFCC similarity detection method is better than traditional short-term energy method. Compared with Euclidean distance measure method, correlation coefficient is better.