Voice Activity Detector (VAD) based on long-term phonetic features
Andrey E. Barabanov, Daniil Kocharov, Sergey I. Salishev, Pavel A. Skrelin, Mikhail Moiseev · ExLing Conferences · 2016
We propose a VAD using long-term phonetically motivated features with auditory masking, and pre-trained decision tree based classifier, which allows capturing syllable level structure of speech and discriminating it from common noise types. The algorithm demonstrates on test dataset almost 100% acceptance of clear voice for English, Chinese, Russian, and Polish speech and 100% rejection of stationary noises independently of loudness with low computational cost.