In this dissertation, various aspects of statistical voice/data multiplexing have been studied, and new results have been obtained in three areas that attract recent research interests.
First, the on-off speech statistics of conversational speech have been measured using an accurate speech detector and a large data base. Based on the measurement results, probability density functions of silence and talkspurt durations have been modeled approximately by two weighted geometric probability density functions. Then, for any value of hangover or fill-in, important speech parameters such as speech activity, and average silence and talkspurt durations have been calculated based on the model of probability density function of silence durations, and compared with those measured. Directly measured values of speech parameters and those calculated have been shown to be in good agreement for a practical range of hangover and fill-in time. For a large hangover time greater than 200 ms, silences and talkspurts have been fit by an exponential distrubution and a constant-plus-exponential distribution in the continuous time domain, respectively. On the other hand, for a large fill-in time greater than 200 ms, silences and talkspurts have been modeled by a constant-plus-exponential distribution and an exponential distribution, respectively. With both large hangover and fill-in values, the talkspurt model agrees closely with the measured data, but the silence model does not agree as closely as the talkspurt model.
Second, the performances of embedded adaptive differential pulse code modulation (ADPCM), embedded adaptive delta modulation (ADM), and the same systems with a delayed-decision scheme have been studied with real speech over a wide dynamic range. The embedded ADPCM and ADM coders have been obtained by modifying the conventional ADPCM and ADM coders. The basic scheme of the embedded ADPCM coder is based on the ADPCM originally proposed by Cummiskey et al. For embedd...