Foreign versus American speaker identification using majority formant minimum value decision criterion
Harbhajan S. Hayre, B. M. Patel · The Journal of the Acoustical Society of America · 1974
Five words were selected to be spoken by ten speakers. Each word was spoken four times by each speaker, and the first three of these utterances were processed and averaged to generate the comparison with the reference formant data obtained from the fourth utterance. The audio data were recorded on a magnetic tape recorder, sampled, and Fourier spectrum analyzed on an IBM 360/44-HSI-SS 100 Hybrid Computer. The resulting power spectra were quantized using five quanta levels, and one-half of the quantum level was used as a threshold. The total power under each formant was divided by its peak level in order to determine its equivalent frequency extent. The reference formant level (a) was used to calculate the second moment of the data formant (b) about the reference level. This technique was employed only in the case when two or more formants (a majority criterion) overlapped, and then 6a minimum value decision criterion was used to determine identification, where all reference formants are used to determine these moments. Preliminary analysis of the results using three foreign-born and six American-born males indicates that a South American male is easier to identify as compared to a South-Indian-born person. Furthermore, American females seem to be more easily identifiable than Texan males. Further work in this area is under way.