A Pitch-Synchronous Study of Formants

Abstract

Introduction: Formants are of fundamental importance in voice science. To date, formants have typically been studied using pitch-asynchronous methods, such as linear-prediction analysis. The results are often incomplete (without level), not objective (with frequencies depending on the preset order p), and require many pitch periods of stationary signals. A method that is accurate, complete, reproducible, and widely applicable is needed.

Method: This study presents a pitch-synchronous method for measuring formants. From the waveform of each pitch period, formants are obtained with high reproducibility, including all formant parameters such as central frequency, level, and bandwidth.

Results: The method was tested on 78 utterances of recorded sustained vowels with simultaneously acquired electroglottograph signals, segmented into 4730 individual pitch periods. For each waveform segment, Fourier analysis was applied to obtain an amplitude spectrum. Formants with three parameters were obtained from each amplitude spectrum. Using these formants, the voice waveforms were regenerated showing strong similarity to the original waveforms. The spectra can be averaged over many pitch periods to reduce noise and to estimate standard deviation.

Conclusions: Measuring formants from the waveform in each pitch period yields accurate, complete, and reproducible results. The method is applicable to live voices, including both speech and singing signals. The results can be used for voice research, speech and singing synthesis, and a quantitative study of phonetics.

Keywords: Voice—Speech—Singing—Vowels—Formants—Timbre.