Automatic accent identification as an analytical tool for accent robust automatic speech recognition
- Submitting institution
-
The University of Birmingham
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 91996222
- Type
- D - Journal article
- DOI
-
10.1016/j.specom.2020.05.003
- Title of journal
- Speech Communication
- Article number
- -
- First page
- 44
- Volume
- 122
- Issue
- -
- ISSN
- 0167-6393
- Open access status
- Access exception
- Month of publication
- June
- Year of publication
- 2020
- URL
-
-
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
1
- Research group(s)
-
-
- Citation count
- 0
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- Accent remains a major cause of errors in automatic speech recognition (ASR) and an obstacle to successful unbiased universal commercial deployment. The paper presents the first comprehensive demonstration of robust ASR for regional accents of British English, using modern deep neural network-based speech models, together with an innovative application of automatic accent identification as a principled tool for selecting training material for accent robust ASR. The impact is commercial, through wider unbiased deployment of computer speech recognition, and academic, through the introduction of a principled approach to training data selection and progress towards novel computationally useful models of accent.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -