Recurrent Neural Network Language Model Adaptation for Multi-Genre Broadcast Speech Recognition and Alignment
- Submitting institution
-
The University of Sheffield
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 2608
- Type
- D - Journal article
- DOI
-
10.1109/TASLP.2018.2888814
- Title of journal
- IEEE/ACM Transactions on Audio, Speech, and Language Processing
- Article number
- -
- First page
- 572
- Volume
- 27
- Issue
- 3
- ISSN
- 2329-9290
- Open access status
- Compliant
- Month of publication
- December
- Year of publication
- 2018
- 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
-
4
- Research group(s)
-
G - Speech and Hearing
- Citation count
- 5
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- Adaptation of Language models is generally difficult. This paper presents an approach that even works with modest amounts of data to adjust models to the target. The approach was used for neural machine translation, speech recognition and alignment. This work was used in several highly cited papers (e.g. https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=7404854 GS 24 , http://eprints.whiterose.ac.uk/117472/8/1598.PDF GS 18, http://eprints.whiterose.ac.uk/109282/1/deena_is16.pdf GS 24) including a very competitive alignment system in the MGB challenge (http://www.mgb-challenge.org/). The methodology is also used in the Voicebase Centre (https://www.voicebase.com/voicebase-speech-technology-partner-university-of-sheffield-to-add-60-phds-to-become-leading-centre-for-ai-speech-technologies/).
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -