Talkspurt

Abstract: Equilibrium point analysis is used to evaluate system behavior in a packet reservation multiple access (PRMA) protocol based network. The authors derive the probability of packet dropping given the number of simultaneous conversations. The authors establish conditions for system stability and efficiency. Numerical calculations based on the theory show close agreement with computer simulations. They also provide valuable guides to system design. Because PRMA is a statistical multiplexer, the channel becomes congested when too many terminals are active. For a particular example it is shown that speech activity detection permits 37 speech terminals to share a PRMA channel with 20 slots per frame, with a packet dropping probability of less than 1%. >

Abstract: The paper describes a voice activity detector (VAD) that can operate reliably in SNRs down to 0 dB and detect most speech at −5 dB. The detector applies a least-squares periodicity estimator to the input signal, and triggers when a significant amount of periodicity is found. It does not aim to find the exact talkspurt boundaries and, consequently, is most suited to speech-logging applications where it is easy to include a small margin to allow for any missed speech. The paper discusses the problem of false triggering on nonspeech periodic signals and shows how robustness to these signals can be achieved with suitable preprocessing and postprocessing.

Abstract: A multiplex transmission system is disclosed in which speech activity is detected as a sequence of talkspurts and intervening silent intervals. The speech information is encoded and accumulated in a buffer store by omitting the silent intervals. A time stamp is associated with each talkspurt code burst to permit approximate reconstruction of the talkspurt time structure at the receiver. The talkspurt code bursts are assembled into packets of optimum size for transmission on shared transmission facilities, such as a Time Assignment Speech Interpolation (TASI) system. The assembled talkspurt packets can be encoded by an adaptive technique responsive to the loading on the transmission facilities and can be transmitted at a rate faster than real-time speech generation to accommodate high-level usage on the shared transmission facilities.

Abstract: This paper deals with the measurement and calculation of various speech temporal parameters of interest in an environment where speech activity detection is employed. In particular it is shown that, based on either a measurement or model of the probability density function (pdf) for silence durations for the case of zero talkspurt "hangover" or "fill-in," that the following temporal parameters can be computed for any value of hangover or fill-in: the mean (and pdf) for silence durations, the mean talkspurt duration, the mean talkspurt rate, and the speech activity. Directly measured values of these parameters and those computed from both measured and fitted versions of the pdf for silence durations are compared and are shown to be in reasonable agreement. The illustrated results are based on measurements of about two minutes of taped male monolog source speech. However, the approach to calculating the above parameters is general in the sense that it can be applied to any measured or modeled pdf for silence durations. The significance of this work lies in the important role that talkspurt hangover plays, for example, in minimizing speech detector induced back-end clipping of talkspurts, reducing exposure to the variable talkspurt delay impairment, and in determining signaling overhead and resource occupancy in various speech interpolation, packet voice, and integrated voice/data systems.