Access to News Audio User Interaction in Speech Retrieval Systems
Download
Report
Transcript Access to News Audio User Interaction in Speech Retrieval Systems
Access to News Audio
User Interaction in Speech Retrieval Systems
by
Jinmook Kim and Douglas W. Oard
May 31, 2002
19th Annual Symposium and Open House
Human-Computer Interaction Lab
University of Maryland
Research Questions
1. What relevance criteria do searchers
rely on when selecting spoken word
materials?
2. What attributes of the recordings do
searchers use as a basis for assessing
relevance?
Search Systems
• NPR Online
Manually prepared transcripts
Human cataloging
• SpeechBot
Automatic Speech Recognition
Automatic indexing
NPR Online
SpeechBot
Qualitative Study Design
• 5 students in a graduate seminar on
Visual and Sound Materials
• Searched both systems
3 started with NPR, 2 with SpeechBot
• Three search topics
Two given by the investigator
One developed by the participant
Data Collection
• Observation
• Think-aloud
• Semi-structured interview
Findings: Relevance Criteria
NPR Online
Topicality
Time range
Type
Novelty
Recency
Listening time
Place
Authority
SpeechBot
Topicality
Time range
Novelty
Type
Accessibility
Findings: Attributes
NPR Online
SpeechBot
Story title
Story summary
Detailed story summary
Date
Audio
Speaker name
Program title
Story length
Mention of a location
Speaker’s affiliation
Extract from transcript
Longer extract from transcript
Audio
Date
Highlighted terms in transcript
Program title
Some Takeaway Messages
• Recognition errors may not bother the
system, but they do bother the user!
• Segment-level indexing can help to
provide effective access
http://www.glue.umd.edu/~jinmook