The Models of Voice Signals’ „Voice” and „Silence” Intervals and their Use

Abstract

There is given an analysis on the models of voice signals’ “voice-silence” intervals. These models are important for creating the systems of modern radio systems. After examining the well-known models, it was discovered that the durations of “voice-silence” intervals are interpreted as independent, the distribution of these durations are often approximated by exponential and that the parameters of these durations depend a lot on the voice activity detectors. These parameters at different sources differ from each other very much. Based on the mentioned above, the conclusion was made that at this moment there are no characteristic models of voice activity that could be unanimously confirmed. The results of this examination confirm the presumptions regarding the duration distribution of exponential voice and silence intervals, when the small talking groups are involved. Also it can be contended that the durations of near-by voice-silence intervals are correlated, though this correlation is not very considerable. The theoretical generalizations regarding more complex voice activity detections methods, which can be substantiated by analysis on specific voice features, are based on these grounds. The elimination of silence intervals goes well with technologies of packet radio connection. However, it is not recommended to expect that this method would markedly increase the amount of at the same time transmitted voice signals in WLAN networks, where amount of speaking voices at the same time is not numerous. The requirements of quality level stipulate by allowing the channel released after excluding the intervals of silence, to the transmission of other data. Ill. 7, bibl. 10 (in English; summaries in Lithuanian, English, Russian).