A Source-Filter Model-Based Unvoiced Speech Detector for Speech Coding
Qian Wang, Xin Du, Weikang Gu · Proceedings of the 2nd International Conference on Computer Science and Electronics Engineering (ICCSEE 2013) · 2013
A novel and easy to realize approach for Voice Activity Detection (VAD) is proposed based on the source-filter speech model in the application of Linear-Prediction-structure speech coding.Generalized Likelihood Ratio Test (GLRT) is adopted to formulate the voice activity detector which contains an unvoiced speech detector and a voiced speech detector respectively.By exploiting the linear predictive analysis coefficients and the pitch information produced in the speech coder, the two separate detectors work effectively in uncorrelated noise without increasing computational complexity significantly.Experimental results show that the unvoiced speech detector outperforms conventional algorithm used in speech coding under various noisy conditions.