Skip to Main Content
Compressed speech through linear predictive analysis and prestored in IC enhances the marketability of speech synthesis. An investigation is done upon the characteristics of the LPC reflection coefficients and reveals a possibility of further compressing the data rate by using an efficient parameter encoding and optimized frame repeat. An improved cost functional utilizing a Log Area Ratio distance measure is employed in this approach, along with an energy weighting function. A dynamic programming is used to maximize the frame length while maintaining spectral distortions within pre-determined thresholds. Good-quality speech, implemented using TI's TMS 5100 or TMS 5200, can be generated from this analysis at 800-850 bits per second. A demonstration of the speech quality will be presented.