JOINT AMI/PASCAL/IM2/M4 Workshop on Multimodal Interaction and Related Machine Learning Algorithms

Confidence Measures in Speech Recognition

author: Stephen Cox, University of East Anglia

Description

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 phrase, word, phone) is correct. Confidence measures are extremely useful in any speech application that involves a dialogue, because they can guide the system towards a more intelligent dialogue that is faster and less frustrating for the user.

Categories

Top: Computer Science: Speech Analysis

You might be experiencing some problems with Your Video player.

*Slides*
0:01	Confidence Measures in Speech Recognition
0:39	Talk Outline
1:11	Why Confidence Measures?
2:35	Previous Work
3:18	Example: Number of hypothesized word-ends as a confidence measure
4:28	PART 1: A General Approach I
5:03	A General Approach II
6:01	Use of a parallel phone recogniser
7:12	Pre-processing for phone correlation
7:30	Phone correlation: distance measure
7:59	Phone correlation: likelihood ratio
8:46	Hypothesising words from phone strings
9:53	LexList: Constructing hypothetical word-sequences
11:58	Hypotheses Made by a Sliding Window of Length 3 Phonemes
12:02	MetaModels—candidate word lists built using phoneme confusions
13:13	Building a Set of Metamodels
15:03	Obtaining a confidence measure from a set of metamodels
15:53	Data and Models
16:30	Performance measurement
18:08	Baseline: “N-best” Confidence
19:05	Performance comparison
19:53	PART II: Use of semantic information in confidence measures
22:05	Preliminary Experiment
23:19	Latent Semantic Analysis
24:31	Co-occurrence matrix W
25:01	Singular Value Decomposition of W
25:56	Data and Representation
26:58	Semantic Score Distributions for Four Words
28:17	Confidence measures from LSA
28:57	Use of a Stop List
29:38	Distribution of PSS scores
30:50	Discussion I
31:23	Discussion II
32:02	Performance of semantic CM
33:43	Final Comments

Lecture rating

People found this lecture:
Worth seeing
because it is:
Valuable and informative
Well presented
Easily understandable
Acceptably recorded
You need to login to cast your vote.

Report a problem or upload files

If you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.