Digital Speech: Coding for Low Bit Rate Communication Systems, 2nd Edition

10

Voice Activity Detection

10.1 Introduction

In voice communications, speech can be characterized as a discontinuous medium because of the pauses which are a unique feature compared to other multimedia signals, such as video, audio and data. The regions where voice information exists are classified as voice-active and the pauses between talk-spurts are called voice-inactive or silence regions. An example illustrating active and inactive voice regions for a speech signal is shown in Figure 10.1.

A voice activity detector (VAD) is an algorithm employed to detect the active and inactive regions of speech. When inactive regions are detected, transmission is generally stopped and only a general description of the background information is transmitted. At the decoder end, inactive frames are then reconstructed by means of comfort noise generation (CNG), which gives natural background sounds with smooth transitions from talk-spurts to pauses and vice versa. To enhance the naturalness of the generated background signal, regular updates of the average information on the background signal (especially necessary during noisy communication environments) is transmitted by the comfort noise insertion (CNI) module of the encoder. The overall structure of the silence compression scheme employing a VAD, CNG, and CNI is shown in Figure 10.2.

Speech communication systems which operate a VAD for compression of inactive speech regions provide various benefits especially useful for bandwidth-limited ...