Speech signal processing using optical method

Abstract

A method has been proposed for the use of optical processing techniques in the analysis and recognition of speech signals. It was realized as an optical processor consisting of a Helium-Neon laser, optical lenses, photographic film plates and diffusers. A personal computer system was further introduced for the total system of optical processing to manage the experimental data. Since the optical processing method has an inherent advantage for high-speed and parallel processing of two-dimensional patterns, the frequency-time pattern of a one-dimensional signal can be obtained without shifting a window along the time axis and the template matching for speech recognition can be conducted in a short period. Utterances of vowels and syllables were analyzed using the processor and the results showed a close aggrement with those obtained by the computer simulation. Nonlinear warping of the time axis, indispensable for spoken word recognition, was shown to be accomplished by controlling the transmittance function of the windowing plate. Template matching of vowel sounds gave a correct recognition of the Japanese five vowels. These results indicate the validity of the optical processor. The use of a liquid crystal plate was also proposed as a windowing plate, and an experiment was conducted on the analysis of vowel sounds. A near-real-time template matching with non-linear time warping is possible by electrically controlling the liquid crystal plate using the result of template matching as a feedback signal.