Meta-Models for Confidence Estimation in Speech Recognition

S. Dasmahapatra; S. J. Cox

doi:10.1109/ICASSP.2000.862107

Meta-Models for Confidence Estimation in Speech Recognition

S. Dasmahapatra
, S. J. Cox

School of Computing Sciences

Research output: Contribution to conference › Paper

3 Citations (Scopus)

Abstract

We describe an approach to confidence estimation that attempts to decouple the contributions of the acoustic and language model components to speech recognition output. The output of the acoustic models when decoding phonemes is itself modelled using HMMs to produce a set of models which we term meta-models. When benchmarked against a “standard” method for assigning confidence (the N-best score), the meta-models gave a relative improvement of 6.2%. Furthermore, it appears that the N-best and meta-models techniques are complementary, because they tend to fail on different words

Original language	English
Pages	1815-1818
Number of pages	4
DOIs	https://doi.org/10.1109/ICASSP.2000.862107
Publication status	Published - Jun 2000
Event	IEEE Conference on Acoustics, Speech and Signal Processing (ICASSP '00) - Istanbul, Turkey Duration: 5 Jun 2000 → 9 Jun 2000

Conference

Conference	IEEE Conference on Acoustics, Speech and Signal Processing (ICASSP '00)
Country/Territory	Turkey
City	Istanbul
Period	5/06/00 → 9/06/00

Access to Document

10.1109/ICASSP.2000.862107

Cite this

@conference{a00256ddb5f34c2ba5cbd97ca92933bc,

title = "Meta-Models for Confidence Estimation in Speech Recognition",

abstract = "We describe an approach to confidence estimation that attempts to decouple the contributions of the acoustic and language model components to speech recognition output. The output of the acoustic models when decoding phonemes is itself modelled using HMMs to produce a set of models which we term meta-models. When benchmarked against a “standard” method for assigning confidence (the N-best score), the meta-models gave a relative improvement of 6.2\%. Furthermore, it appears that the N-best and meta-models techniques are complementary, because they tend to fail on different words",

author = "S. Dasmahapatra and Cox, \{S. J.\}",

year = "2000",

month = jun,

doi = "10.1109/ICASSP.2000.862107",

language = "English",

pages = "1815--1818",

note = "IEEE Conference on Acoustics, Speech and Signal Processing (ICASSP '00) ; Conference date: 05-06-2000 Through 09-06-2000",

}

TY - CONF

T1 - Meta-Models for Confidence Estimation in Speech Recognition

AU - Dasmahapatra, S.

AU - Cox, S. J.

PY - 2000/6

Y1 - 2000/6

N2 - We describe an approach to confidence estimation that attempts to decouple the contributions of the acoustic and language model components to speech recognition output. The output of the acoustic models when decoding phonemes is itself modelled using HMMs to produce a set of models which we term meta-models. When benchmarked against a “standard” method for assigning confidence (the N-best score), the meta-models gave a relative improvement of 6.2%. Furthermore, it appears that the N-best and meta-models techniques are complementary, because they tend to fail on different words

AB - We describe an approach to confidence estimation that attempts to decouple the contributions of the acoustic and language model components to speech recognition output. The output of the acoustic models when decoding phonemes is itself modelled using HMMs to produce a set of models which we term meta-models. When benchmarked against a “standard” method for assigning confidence (the N-best score), the meta-models gave a relative improvement of 6.2%. Furthermore, it appears that the N-best and meta-models techniques are complementary, because they tend to fail on different words

U2 - 10.1109/ICASSP.2000.862107

DO - 10.1109/ICASSP.2000.862107

M3 - Paper

SP - 1815

EP - 1818

T2 - IEEE Conference on Acoustics, Speech and Signal Processing (ICASSP '00)

Y2 - 5 June 2000 through 9 June 2000

ER -