Journal of Engineering and Applied Sciences
ISSN: Online 1818-7803ISSN: Print 1816-949x
Abstract
This study reports on recent improvements made to a Phoneme-Matching Algorithm (PMA) reported in a previous study. Similar to its predecessor, the purpose of the Enhanced PMA (EPMA) is to identify word recognition errors in automatically generated transcripts detailing the speech content of digital multimedia soundtracks that are routinely queried by professional researchers (such as academics and archivists). In order to alert a user to the possibility that a particular search term may have been incorrectly recognised as some other word or phrase, the EPMA when invoked during a query operation will parse the transcript’s text to locate words or phrases of similar phonetic structure to the query term and then present these suspected speech recognition errors to the user for consideration. The EPMA’sperformance has been improved by incorporating techniques to learn from user feedback concerning error identification. When tested on a corpus of digital multimedia, the EPMA averaged an 80.55% success rate in correctly identifying words/phrases which were actually instances of misrecognised query terms.
How to cite this article:
James Carmichael. An Enhanced Phoneme-Matching Algorithm Enhanced by User Feedback to Identify Possible Automatic Speech Recognition Transcription Errors.
DOI: https://doi.org/10.36478/jeasci.2014.457.463
URL: https://www.makhillpublications.co/view-article/1816-949x/jeasci.2014.457.463