A study of voice/non-voice discrimination method using neural networks for integrated packet switching system

摘要

Voice packet technology, which handles analog signals such as DTMF (dual-tone multifrequency) signals and voiceband data as well as voice, image and data, is a key technology for a fast packet-switching system for multimedia communication. Transmission of voice signal in packets involves echo canceling, bandwidth-compression coding, and silence compression. However, if nonvoice signals like DTMF signals and voiceband data undergo such processing, communication quality may deteriorate due to malfunction or erroneous detection at the receiving side. The authors have developed a voice/nonvoice discrimination method by combining digital signal processing and a neural network. They confirmed that the method showed good characteristics by simulation and by running it on a DSP.