Speech compression: a functional approximation approach

Abstract

This paper presents research undertaken in the field of speech compression with a low cost speech processing system developed around an APPLE II microcomputer. Unlike some of the more popular techniques of speech compression based on statistical analysis of the speech waveforms in the frequency domain, speech compression was approached by the authors with the perspective of functional approximations of speech waveforms in the time domain. These functional approximations would be evaluated in a parallel processing mode. Research activities discussed in this paper include: the design and implementation of a parallel interface between the APPLE II and an analog to digital converter; the development of machine language programs to accurately sample sound waveforms; the development of software for the analysis of converted data, including a real time scrolling graphics routine and also a graphics hardcopy routine which controls a dot-matrix printer; the development of the compression technique utilizing an intelligent sampling routine; and the design of a pipelined parallel processing network of microprocessors, to evaluate in real time a polynomial function representing a speech waveform. The compression technique developed currently allows four minutes of good quality speech to be stored on one floppy disk.