## Schlüter: ‘Using Word Probabilities as Confidence Measures (1998)

Venue: | in Proc. IEEE Int. Conf. on Acoustics, Speech and Signal Processing |

Citations: | 49 - 7 self |

### BibTeX

@INPROCEEDINGS{Wessel98schlüter:‘using,

author = {Frank Wessel and Klaus Macherey and Ralf Schlüter},

title = {Schlüter: ‘Using Word Probabilities as Confidence Measures},

booktitle = {in Proc. IEEE Int. Conf. on Acoustics, Speech and Signal Processing},

year = {1998},

pages = {225--228}

}

### Years of Citing Articles

### OpenURL

### Abstract

Estimates of confidence for the output of a speech recognition system can be used in many practical applications of speech recognition technology. They can be employed for detecting possible errors and can help to avoid undesirable verification turns in automatic inquiry systems. In this paper we propose to estimate the confidence in a hypothesized word as its posterior probability, given all acoustic feature vectors of the speaker utterance. The basic idea of our approach is to estimate the posterior word probabilities as the sum of all word hypothesis probabilities which represent the occurrence of the same word in more or less the same segment of time. The word hypothesis probabilities are approximated by paths in a wordgraph and are computed using a simplified forward-backward algorithm. We present experimental results on the NORTH AMERICAN BUSINESS (NAB’94) and the German VERBMOBIL recognition task. 1.

### Citations

85 |
A word graph algorithm for large vocabulary, continuous speech recognition
- Ney, Aubert
- 1994
(Show Context)
Citation Context ...Wa = (a1 :::a N ) is therefore computed as: p(Wa) = p(a1) \Delta m\Gamma1 Y i=2 p(a i ja i\Gamma1 1 ) \Delta N Y i=m p(a i ja i\Gamma1 i\Gammam+1 ) : (4) In our speech recognition system a word graph =-=[5]-=- is a directed graph whose nodes are interpreted as starting and ending times of word hypotheses and whose edges represent word hypotheses. The acoustic probabilities p(x t e t a jw) are therefore sto... |

74 | Estimating confidence using word lattices
- Kemp, Schaaf
(Show Context)
Citation Context ...measures has either investigated the computation of confidence measures during the acoustic decoding process, e.g. [1, 2] or the computation of confidence measures on the basis of word lattices, e.g. =-=[4]-=- and n-best lists, e.g. [6]. Gillick et at. [3] have estimated and evaluated their confidence measure in the framework of a probabilistic approach, making use of generalized linear models for relating... |

48 |
Confidence Measures for the Switchboard Database
- Cox, Rose
- 1996
(Show Context)
Citation Context ... and the dialogue duration can be drastically shortened. Previous work on confidence measures has either investigated the computation of confidence measures during the acoustic decoding process, e.g. =-=[1, 2]-=- or the computation of confidence measures on the basis of word lattices, e.g. [4] and n-best lists, e.g. [6]. Gillick et at. [3] have estimated and evaluated their confidence measure in the framework... |

48 | Neural network based measures of confidence for word recognition
- Weintraub, Beaufays, et al.
- 1997
(Show Context)
Citation Context ...re in the framework of a probabilistic approach, making use of generalized linear models for relating a confidence feature vector directly to the probability of a word to be correct. Weintraub et at. =-=[9]-=- have used artifical neural networks to model the relation between the different features and this probability. The computation of posterior word probabilities in this paper can be seen as an extensio... |

47 | Confidence measure for spontaneous speech recognition - Schaaf, Kemp |

27 | and acoustic confidence annotation for large vocabulary speech recognition
- Chase, “Word
- 1997
(Show Context)
Citation Context ... and the dialogue duration can be drastically shortened. Previous work on confidence measures has either investigated the computation of confidence measures during the acoustic decoding process, e.g. =-=[1, 2]-=- or the computation of confidence measures on the basis of word lattices, e.g. [4] and n-best lists, e.g. [6]. Gillick et at. [3] have estimated and evaluated their confidence measure in the framework... |

24 |
Improved estimation, evaluation and applications of confidence measures for speech recognition
- Sui, Gish, et al.
- 1997
(Show Context)
Citation Context ...CjX) can be interpreted as the entropy of the tag sequence attached to the recognizer output, provided with the information contained in the confidence measure. For details, the reader is referred to =-=[8]-=-. Although S can easily be interpreted as the relative reduction in entropy, it is no longer sensibly defined as soon as the posterior probability for a word to be correct equals one, even though the ... |

7 |
A probabilistic approach to confidence measure and evaluation
- Gillick, Ito, et al.
(Show Context)
Citation Context ...n of confidence measures during the acoustic decoding process, e.g. [1, 2] or the computation of confidence measures on the basis of word lattices, e.g. [4] and n-best lists, e.g. [6]. Gillick et at. =-=[3]-=- have estimated and evaluated their confidence measure in the framework of a probabilistic approach, making use of generalized linear models for relating a confidence feature vector directly to the pr... |

5 |
Stolcke: ‘Neural-Network Based Measures of Confidence for Word Recognition
- Weintraub, Beaufays, et al.
- 1997
(Show Context)
Citation Context ...re in the framework of a probabilistic approach, making use of generalized linear models for relating a confidence feature vector directly to the probability of a word to be correct. Weintraub et at. =-=[9]-=- have used artifical neural networks to model the relation between the different features and this probability. The computation of posterior word probabilities in this paper can be seen as an extensio... |

4 |
Obtaining confidence measures from sentence probabilities". EUROSPEECH-1997: p
- Bernhard
- 1997
(Show Context)
Citation Context ...gated the computation of confidence measures during the acoustic decoding process, e.g. [1, 2] or the computation of confidence measures on the basis of word lattices, e.g. [4] and n-best lists, e.g. =-=[6]-=-. Gillick et at. [3] have estimated and evaluated their confidence measure in the framework of a probabilistic approach, making use of generalized linear models for relating a confidence feature vecto... |

4 |
and Acoustic Confidence Annotation for Large Vocabulary Speech Recogniton
- ‘Word
- 1997
(Show Context)
Citation Context ... and the dialogue duration can be drastically shortened. Previous work on confidence measures has either investigated the computation of confidence measures during the acoustic decoding process, e.g. =-=[1, 2]-=- or the computation of confidence measures on the basis of word lattices, e.g. [4] and n-best lists, e.g. [6]. Gillick et at. [3] have estimated and evaluated their confidence measure in the framework... |