VOICE ACTIVITY DETECTION (VAD) USING BIPOLAR PULSE ACTIVE (BPA) FEATURES

Abstract

This paper introduces a new method of extracting features of a signal called Bipolar Pulse Active (BPA) features. The BPA extracts information from signals and decomposes them, using a series of harmonically related periodic triangular waveforms, into a finite set of Pulse Domain features. The BPA features contain information such as the rate of change between amplitudes, durations and frequencies within the original signal. The transformed signals can be further processed for some desirable application. In this paper, BPA feature is modified to describe its energy spectrum to be used in the application of Voice Activity Detection (VAD) of a speech signal