An Improved Speech Endpoint Detection Method Based on a Gaussian-Gamma Model
Hui Liu · Microelectronics & Computer · 2008
Speech endpoint was an important segment in the speech preprocess.Some traditional endpoint detection methods were ineffective in real noisy environments.To improve the robustness of the vad,we proposed the improved algorithm that based on a Gaussian-gamma model in the DCT domain.The distribution of clean speech was assumed to be Gamma.The parameter was estimated by maximum likelihood(ML)and predicted estimation approaches.This algorithm improved the precision and robustness of the endpoint detection.The simulation experiment demonstrated that the algorithm had a better robustness.It had a good performance even in low SNR environments.