Do listeners rely on dynamic spectral properties in the recognition of high-pitched vowels?

Abstract

Previous research has demonstrated that intrinsic spectral changes play a significant role in vowel perception. It is possible that dynamic properties are of particular importance at high fundamental frequencies (f o) where the transfer function of the vocal tract is severely undersampled. The purpose of the present work was to determine whether the phonological function of vowels with very high f os can be preserved if listeners' identification is based exclusively on static spectral cues. Harmonic additive synthesis was used to generate 324 steady-state versions of the vowels /i y e o ɛ a o u/ with nine different f os between 220 and 880 Hz (all 250 ms). The gross spectral shape of each synthesized sound corresponded to a smoothed spectrum derived from a respective template vowel that was naturally produced in isolation by a female native German talker. Native German listeners either identified the spoken or re-synthesized isolated vowels in a two-alternative choice task. Preliminary data showed that listeners' sensitivity (A′) was above chance level in both conditions at all f os, but significantly higher in the spoken vowel condition. This supports the prediction that dynamic properties enhance listeners' identification at very high f os, but also indicates that phonological distinctiveness is to some degree preserved at high f os even when no dynamic spectral information is present. [Work supported by the Swiss National Science Foundation (SNSF) grants P2ZHP1_168375168375 and P400PG_180693180693.]Previous research has demonstrated that intrinsic spectral changes play a significant role in vowel perception. It is possible that dynamic properties are of particular importance at high fundamental frequencies (f o) where the transfer function of the vocal tract is severely undersampled. The purpose of the present work was to determine whether the phonological function of vowels with very high f os can be preserved if listeners' identification is based exclusively on static spectral cues. Harmonic additive synthesis was used to generate 324 steady-state versions of the vowels /i y e o ɛ a o u/ with nine different f os between 220 and 880 Hz (all 250 ms). The gross spectral shape of each synthesized sound corresponded to a smoothed spectrum derived from a respective template vowel that was naturally produced in isolation by a female native German talker. Native German listeners either identified the spoken or re-synthesized isolated vowels in a two-alternative choice task. Preliminary data showed that li...