A practical method of estimating the time-varying degree of vowel nasalization from acoustic features

This paper presents a simple and easy-to-use method of creating time-varying signal the degree nasalization in vowels, generated from acoustic features measured oral nasalized vowel contexts. The is presented for separate models constructed using two sets features: (1) an uninformed set 13 Mel-frequency cepstral coefficients (MFCCs) (2) combination MFCCs phonetically informed 20 nasality derived previous research. Both are compared against traditional approaches to estimating acoustics: A1-P0 A1-P1, as well their formant-compensated counterparts. Data include productions six speakers different language backgrounds, producing 11 qualities within quadrilateral. results each methods nasometric measurements, representing objective “ground truth” nasalization. suggest that proposed more robust than conventional approaches, generating signals which correlate strongly with measures across all accurately approximate change Finally, experimental example provided help researchers implement own study designs.