Frequency and Time-Domain Differentiation of Speech Contour Data for Voice Recognition
Donald L. Snider, Harry W. Mergler · The Journal of the Acoustical Society of America · 1970
In an attempt to enhance voice recognition capabilities, several transformations have been applied to the speech acoustical signal. The transformations include first- and second-order partial differentiation of the amplitude of the frequency-time-amplitude contour with respect to frequency and with respect to time, and variable segmentation of this contour and the derivative contours based upon criteria obtained from the partial derivative information. To provide the initial data base for the experiments, an instrumentation system was implemented to digitize the speech contours with the high degree of resolution of 256 frequency channels and 125 time increments or 32 000 data points. Computer-generated contour spectrograms were produced for the original contour and for the first and second partial-derivative contours. Speech recognition and speaker identification systems have been computer simulated to evaluate the effectiveness of the transformations.