Voice Activity Detection (VAD) is a (software) module to detect if there is a silent period during a speech connection. This knowledge is used to switch off the transmission during silent periods. This is called discontinuous transmission (DTX).

Voice activity detection must be very accurate. If it considers speech as noise, part of the conversation will get lost. If on the other hand, it can not effectively detect a silent period, DTX will not be very effective.

If the transmission stops during silent periods, the background acoustic noise abruptly disappears at the receiving end. This can be very annoying for the receiving party. To overcome this problem 'comfort noise' is generated at the receiving end whenever the transmission is switched off. The comfort noise is generated by a Comfort Noise Generator (CNG). To generate 'realistic' acoustic background noise the DTX module sends Silence Insertion Descriptor (SID) frames periodically to the CNG module at the receiving end.

See also