How the peak glottal area affects linear predictive coding-based formant estimates of vowels.

Area of Science:

• Acoustic Phonetics
• Speech Production
• Bioacoustics

Background:

• Linear Predictive Coding (LPC) is standard for estimating formant frequencies from speech.
• LPC assumes a closed glottis, but real vocal fold vibration involves a time-varying glottal area.
• Previous research suggests LPC formants may differ from actual vocal tract resonances with a dynamic glottis.

Purpose of the Study:

• To systematically investigate the deviation between vocal tract resonances (closed glottis) and LPC-estimated formants.
• To quantify the impact of varying peak glottal areas on formant frequency estimation during phonation.

Main Methods:

• Utilized physical vocal tract models with a self-oscillating rubber vocal fold model.
• Employed computer simulations of interacting source and filter models.
• Analyzed ten vocal tract resonators representing different vowels.

Main Results:

• The first formant (F1) frequency increased with the peak glottal area.
• Second (F2) and third (F3) formant frequencies showed no systematic changes with glottal area.
• The effect of glottal area on F1 was most pronounced for close-mid to close vowels.

Conclusions:

• The assumption of a closed glottis in LPC analysis can lead to inaccurate formant frequency estimations, particularly for F1.
• The dynamic nature of the glottis significantly influences the acoustic output, especially for certain vowel qualities.
• Findings highlight the importance of considering glottal dynamics in speech analysis and modeling.