The role of talker change, vowel change and speech perceptual measure in hearing impaired phone recognition

Abavisani, Ali

Permalink

https://hdl.handle.net/2142/116255 Copy

Description

Title

The role of talker change, vowel change and speech perceptual measure in hearing impaired phone recognition

Author(s)

Abavisani, Ali

Issue Date

2022-07-18

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

Hasegawa-Johnson, Mark

Doctoral Committee Chair(s)

Hasegawa-Johnson, Mark

Committee Member(s)

• Smaragdis, Paris
• Oelze, Michael L
• Wickesberg, Robert E

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• signal processing
• speech
• speech recognition
• automatic speech recognition
• hearing aids
• hearing impaired
• hearing
• psycho acoustics

Abstract

In this study, we first investigated the role of frequency-dependent insertion gain for hearing aids on speech recognition of Hearing Impaired (HI) listeners, having mild-to-moderate hearing loss. The outcome indicated that for a subset of HI listeners, amplifying isolated Consonant-Vowel (CV) speech phones to compensate for hearing loss is not the ultimate solution and, in fact, it may degrade their recognition performance. Second, we focused on the perceptual measure of speech intelligibility in background speech-weighted noise, denoted as SNR90, based on Normal Hearing (NH) phone recognition. The SNR90 measure indicates the noise levels in which an NH listener would recognize the CV with at least 90% accuracy. This measure will be helpful in determining proper speech stimuli corpus for a speech-based hearing test. Thus, the third area of focus in this study was to fully investigate the role of SNR90 perceptual measure in terms of enhancement of SNR90 through changing the talker on the presented CV. The results show that enhancement of SNR90 improved phone recognition for all HI listeners. The fourth area of focus was to investigate the role of frequency fine-tuning of the CV by keeping the perceptual measure SNR90 in the same level, but changing the vowel. The results indicated that where the vowel change increased the energy of primary cue region of the consonant, the phone recognition was improved for HI listeners. Finally, as the perceptual measure SNR90 plays a very important role in HI phone recognition, we devised models based on convolutional neural networks to automatically estimate the SNR90 of CV sounds. The models outperform the only available baseline, i.e., estimation of speech intelligibility measure by state of the art automatic speech recognition (ASR) system.

Graduation Semester

2022-08

Type of Resource

Thesis

Copyright and License Information

Copyright 2022 Ali Abavisani

Owning Collections