Abstract
The paper deals with the design of an analyser-coder for construction of a digital code pattern from a human speech input suitable for the use of computers from which the intelligence content can be extracted without any human intervention.
In the first part of this paper the major functions necessary in any pattern recognition system are noted, aspects of decision theory as pertinent to the present work are mentioned and a theoretical model for speech analysis by synthesis is described very briefly to introduce the work.
The second part of the paper describes the experimental part of the system, human speech analyser and matrix coder. Some of the circuitry are conventional, others are special circuits designed for critical applications.
The underlying theme of the work is to design a frequency quantized, continuous, short-term spectrum analysis technique capable of extracting statistically invariant properties of the human speech pattern to a binary quantized representation produced by the measurement apparatus and used as the input data to the categorizer portion of the recognition system.