Voice Activity Detection for Speaker Verification Systems

摘要

Complex algorithm for speech activity detection was presented in this article. It is based on speech enhancement, features extraction and final detection algorithm. The first one was published in ETSI standard as a module of "Advanced front-end feature extraction algorithm" in distributed speech recognition system. It consists of two main parts, noise estimatiom and Wiener filtering. For the final detection modified linear prediction coefficients and spectral entropy features are extracted form denoised signal.