Data rate reduction of gain and pitch parameters in an LPC vocoder

The purpose of this research was to investigate the utility of employing data compression techniques on the gain and pitch parameters needed for synthesis in a pitch-excited LPC vocoder. As the overall data rates required for satisfactory speech synthesis decrease, the relative portion of the information rate required for excitation data increases (in most of the less complex synthesis systems). This study investigated the use of differential quantization techniques on the pitch period and gain data used in an LPC vocoder. In order to include a wide range of quantizer design procedures, four different types of adaptive predictors and four different types of quantizers were considered, thereby allowing realization of sixteen distinct systems. The results of each system were evaluated objectively using a mean squared error criterion, and subjectively via informal listening tests. While the study was not an exhaustive one, the results obtained indicate a modest potential for data rate reduction on the excitation information. More generally, this study confirms the concept that LPC techniques can be readily applied to non-quasi-periodic signals.