Fundamental frequency synchronous spectral analysis for vowel classification
Bingjun Dai, Stephen A. Zahorian · The Journal of the Acoustical Society of America · 1998
A method is described and tested for spectral analysis of vowels using window lengths that are integer multiples of the pitch period for voiced signals. For segments of vowels selected from steady-state vowels, the average fundamental frequency is first determined using ‘‘standard’’ autocorrelation methods. The acoustic signal is then analyzed again to determine fundamental frequency on a cycle-by-cycle basis. The signal is resampled at a higher rate so that each pitch period has the same number of samples. Spectral analysis is implemented such that harmonics of the fundamental coincide with FFT samples. Cepstral analysis is then performed on this spectrum with and without fundamental-frequency-dependent frequency shifting as a speaker normalization. This processing method was tested using a database of 150 speakers (50 children, 50 women, and 50 men) for ten monopthong vowels produced in isolation. As a control, spectral analysis was also performed on Hamming-windowed speech segments without regard to fundamental frequency. Neural network classification tests indicated only minor changes in classification rates with the fundamental-synchronous method versus the conventional method. The implication of this work is that spectral smearing due to window effects is a minor effect for vowel spectral analysis. [Funded by NSF, Grant No. NSF-BES-9411607.]