Inverse filtering of nasalized vowels using synthesized speech.

Area of Science:

• Phonetics
• Speech Science
• Acoustic Analysis

Background:

• Inverse filtering is used to estimate glottal parameters assuming an all-pole vocal tract model.
• Nasal coupling, common in speech, may influence the accuracy of these estimates.

Purpose of the Study:

• To investigate how increased nasal coupling affects glottal parameter estimations derived from inverse filtering.
• To quantify the impact of nasal coupling on specific glottal source parameters (EE, RG, RA, RK, OQ).

Main Methods:

• Synthesized steady-state Swedish vowels using the HLsyn synthesizer.
• Systematically varied nasal aperture area (an) while keeping other parameters constant.
• Applied inverse filtering and fitted the Liljencrants and Fant source model to estimate glottal parameters.

Main Results:

• Nasal coupling significantly affects the RA parameter estimate, a key determinant of spectral slope.
• Estimation errors were highest for high vowels, impacting RA and other parameters.
• Except for RA, parameter estimation errors were relatively small despite varying degrees of nasal coupling.

Conclusions:

• Inverse filtering with an all-pole vocal tract assumption is sensitive to nasal coupling, especially for RA.
• High vowels are particularly susceptible to inaccuracies in glottal parameter estimation when nasalized.
• While RA estimates are notably affected, other glottal parameters show less distortion due to nasal coupling.