Digital circuit techniques for speech analysis

The fundamentals of voice production and hearing are briefly reviewed with respect to the nature of the complex speech waveform. Some early speech devices are noted and current work in the field of computer simulation is considered. A general description of an experimental word code generator developed at the IBM (International Business Machines Corporation) Endicott Development Laboratory is given. This device produces a compressed digital code from the spoken word as a function of frequency, intensity, and time. Individual sound patterns are coded as a function of the changes in the word itself. The code is produced as the word is spoken and is displayed in matrix form for immediate analysis. Readout to a card punch or other storage media is easily accomplished. The operation of the system is described with special reference to the circuit configurations and techniques developed for the word code generator. The advantages and disadvantages of the digital representation of speech events are discussed considering storage economy, information transfer, and pattern recognition. It is concluded that digital techniques for speech analysis should find usage in speech therapy, speech recognition, and speech synthesis.