Accuracy and reliability of formant frequency measurements in synthetic infant CV syllables

Abstract

An investigation of the reliability and accuracy of measurement of infant formant frequencies was conducted. Four factors known to render formant measurement difficult—pitch inflection, glottal noise, jitter, and nasalization—were incorporated into synthesized CV (C = /b,d/, V = /i,a,ɛ/) syllables appropriate to 3-month-olds. The syllables were of high intelligibility and quality. Three fundamental frequency contours (standard rise/fall, rising, flat), three F0 levels (325, 375, 425 Hz), four voice/noise ratios (+20, +5, 0, −15), three frequency jitter rates (250, 500, 750), and one degree of nasal coupling were used. Formant frequency measurements were made in two ways for each token: (1) as single FFT cross sections using a 10-ms window and (2) as averaged, time-advanced FFT spectra. In the second condition, three successive FFTs taken at 2.5-ms rightward shifts of the 10-ms sampling window were averaged to produce the FFT spectrum for measurement. Three observers independently measured F1, F2, and F3 and results were compared to synthesis parameter values. Inter- and intraobserver reliability was also computed. Results are discussed with regard to the influence of consonant place and vowel type upon accuracy in each of the four conditions of waveform perturbation.