On the spectral information content of vowels

Kim, Austin J.

Permalink

https://hdl.handle.net/2142/29782 Copy

Description

Title

On the spectral information content of vowels

Author(s)

Kim, Austin J.

Issue Date

2012-02-06T20:16:31Z

Director of Research (if dissertation) or Advisor (if thesis)

Allen, Jont B.

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

Vowel Perception

Abstract

This thesis explores the human perception of vowels in a consonant-vowel (CV) context by examining the data of two experiments. The first experi- ment, CV06-SWN, examines the effects of speech-weighted noise (SWN) on vowel perception. The vowels examined are /A/, /o/, /u/, /U/, /E/, /æ/, /3~/, and /I/, while the consonants are /d/, /g/, /k/, /p/, /s/, /t/, /S/, /Z/, and /z/. Results demonstrate that there is a threshold signal-to-noise (SNR) ratio at which the perception scores abruptly drop to chance, once the critical cue(s) to perception are masked. The results also show that human percep- tion of vowels in quiet is largely dependent on context, as the error rates of lax vowels are far higher than those of tense vowels. Lax vowels rarely occur in a CV context in English. This seems to be a novel observation. The second experiment, HL11, examines the effects of filtering, using low-pass or high-pass filters at various cut-off frequencies, on vowel perception. In a /hV/ context, HL11 uses eleven vowels: /A/, /O/, /2/, /o/, /u/, /U/, /E/, /æ/, /3~/, /i,/ and /I/. Results from the low-pass filter conditions indicate that for most of the vowels, the second formant (F2) is the crucial feature, as the vowel performance drops dramatically once it has been suppressed. The confusions indicate that listeners are also using the first formant (F1). Results from the high-pass filter conditions indicate the same. For many of the vowels, the vowel performance curve dips at the cut-off frequencies between F1 and F2. The primary confusion therein is a vowel with a higher F1 but similar F2. It is as if the the energy between the suppressed F1 and F2 serves as a “false” F1. The vowel performances of all vowels except /i/ drop dramatically as F2 is suppressed. The primary confusion at the highest cut-off frequency is /i/, since its F2 is similar to the F3 of the other vowels.

Graduation Semester

2011-12

Permalink

http://hdl.handle.net/2142/29782

Copyright and License Information

Copyright 2011 Austin J. Kim

Owning Collections