Confidence Measures in Speech Recognition thumbnail
Pause
Mute
Subtitles
Playback speed
0.25
0.5
0.75
1
1.25
1.5
1.75
2
Full screen

Confidence Measures in Speech Recognition

Published on Feb 25, 20079250 Views

A confidence measure (CM) is a number between 0 and 1 that is applied to speech recognition output. A CM gives an indication of how confident we are that the unit to which it has been applied (e.g. a

Related categories

Chapter list

Confidence Measures in Speech Recognition00:01
Talk Outline00:39
Why Confidence Measures?01:11
Previous Work02:35
Example: Number of hypothesized word-ends as a confidence measure03:18
PART 1: A General Approach I04:28
A General Approach II05:03
Use of a parallel phone recogniser06:01
Pre-processing for phone correlation07:12
Phone correlation: distance measure07:30
Phone correlation: likelihood ratio07:59
Hypothesising words from phone strings08:46
LexList: Constructing hypothetical word-sequences09:53
Hypotheses Made by a Sliding Window of Length 3 Phonemes11:58
MetaModels—candidate word lists built using phoneme confusions12:02
Building a Set of Metamodels13:13
Obtaining a confidence measure from a set of metamodels15:03
Data and Models15:53
Performance measurement16:30
Baseline: “N-best” Confidence18:08
Performance comparison19:05
PART II: Use of semantic information in confidence measures19:53
Preliminary Experiment22:05
Latent Semantic Analysis23:19
Co-occurrence matrix W24:31
Singular Value Decomposition of W25:01
Data and Representation25:56
Semantic Score Distributions for Four Words26:58
Confidence measures from LSA28:17
Use of a Stop List28:57
Distribution of PSS scores29:38
Discussion I30:50
Discussion II31:23
Performance of semantic CM32:02
Final Comments33:43