This thesis is about eh...Finding the optimal combination of filled pauses to improve a keyword spotting automatic speech recogniser.
Keywords
No Thumbnail Available
Authors
Issue Date
2015-08-27
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
This study aims to improve a keyword spotting (KWS) automatic speech recogniser (ASR) system by adding different combinations of filled pauses to its keyword list. The research questions the study revolves around are: “How do filled pauses manifest themselves in spontaneous spoken Dutch telephone conversations?” and “What is the optimal combination of categories of filled pauses in order to improve a LVCSR-based KWS ASR system?”. It appears that filled pauses manifest themselves in 17 different ways, and adding all 17 of those to the keyword list of the ASR system improves its performance the most. The found different types of filled pauses in Dutch were in line with findings of the English and German languages (Rose, 1998; Batliner et al. 1995). The present study provides a new approach to automatic detection of filled pauses based on treating filled pauses as keywords. This approach could be implemented in KWS! ASR syst ems.
Description
Citation
Supervisor
Faculty
Faculteit der Letteren