Abstract
Recognizing the importance and usefulness of analysis-by-synthesis technique in the analysis and recognition of speech, various problems on its application to formant frequency extraction were reexamined in relation to the characteristics of our analyzer. It was found necessary to add a process of pitch frequency adjustment in order to improve the accuracy of extraction and an efficient method of pitch frequency adjustment was developed. This improvement could make the process applicable not only to male voices but also to female voices without any deterioration of results. The over-all accuracy of this method was better than that of the moment calculation method and may approach to the same order of D. L. of formant frequency by human perception. The whole process of automatic extraction of formant frequencies by analysis-by-synthesis technique was programed and tested with the practical input speech spectra. The time scale of the process by NEAC-2203 computer was of the order of 10^5 and unrealistic even if the first good approximations obtained by the moment calculation method was fully employed. The extraction of formant frequencies on the full scale is scheduled to be carried on by the new computer NEAC-2206 in the future.
