Estimating acoustic speech features in low signal-to-noise ratios using a statistical framework
- Submitting institution
-
The University of East Anglia
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 182620634
- Type
- D - Journal article
- DOI
-
10.1016/j.csl.2016.08.001
- Title of journal
- Computer Speech and Language
- Article number
- -
- First page
- 1
- Volume
- 42
- Issue
- -
- ISSN
- 0885-2308
- Open access status
- Compliant
- Month of publication
- March
- Year of publication
- 2017
- 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
- Acoustic speech features, such as voicing class, fundamental frequency and spectral envelope, have been estimated using many different techniques and their accuracy deteriorates as the speech becomes noisy. This paper proposes a single statistical framework from where these acoustic features are estimated. Furthermore, adaptation methods that have been successful in robust speech recognition in noise have been applied and are shown to improve significantly the estimation accuracy of acoustic features at low signal-to-noise ratios, exceeding that of existing methods. An outcome of this work has been the permanent employment at a leading speech processing company of the lead author.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -