A new multi-parameter dual-threshold state discrimination algorithm for voice activity detection
Xinyan Zeng, Guojun Zhao · 2013
The desired results cannot be achieved by applying generally accepted algorithm of voice activity detection (VAD). Against such background, a comprehensive algorithm featuring multiple parameters and dual-threshold state discrimination algorithm is proposed. This algorithm sets both a high and a low threshold based on the fundamental characteristic parameters such as the short-time average energy (STAE), the short-time zero-domain-crossing rate (ZDCR) and the time duration, with reference to the design of finite state method. The interval between these two thresholds helps to detect voice activity with the improved results. After the new algorithm was compared and analyzed through Matlab, it experimentally tested on a DSP-based processing system. It proves that it is more effective and accurate than the widely accepted one.