VoIP Voice and Fax Signal Processing

4.1 VAD/CNG AND Codecs

The speech is a half-duplex under normal conversation, meaning one person at a time speaks into the phone. In a telephone conversation between terminals (or users) A–B, speech from A to B or B to A is present at a given time. In the middle of the same person's speech, significant durations of inactive (silence) zones occur. VAD/CNG exploits inactive zones and eliminates the need for sending voice packets to the network, thus saving about 40% to 50% of Internet bandwidth. The bandwidth saving happens with a certain degradation of voice quality, even though it is not a significant degradation. As a general recommendation, the use of VAD/CNG has to be eliminated on availability of sufficient network bandwidth for a VoIP voice call.

VAD/CNG is embedded with some compression codecs such as G.729AB and G.723.1A. These codecs work based on the vocal tract model. VAD and the comfort noise created during silence closely match and continue to generate the original background modeled through the vocal tract. The codecs G.711, G.726, and G.728 do not have a built-in VAD/CNG scheme. These codecs are used with an external VAD/CNG scheme. Two popular VAD/CNG schemes for G.711 and G.726 exist.

Power- or energy-based VAD/CNG is the simplest VAD/CNG scheme used for G.711 and G.726 codecs. The names “power based” or “energy based” are used for representing this scheme. In this chapter, the power-based VAD/CNG name is used. Some engineers also call this power-based VAD/CNG a ...