Loading...

Computation of Confidence Measure for Detection of out of Vocabulary Words in a Continuous Speech Recognition System

Sakhaee, Elham | 2010

476 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 42089 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Samti, Hossein
  7. Abstract:
  8. Automatic Speech Recognition (ASR) engines are too much sensitive to conditions such as noise, transmission line quality, etc. Thus in any real-world application ASR systems should be able to automatically assess reliability or probability of correctness for every decision made by the systems. One technique to increase intelligence of an ASR system is to compute a score, called “confidence measure” to indicate reliability of any recognition decision made by the system. This score can be computed at any required level such as phonemes, syllabi, words or even the whole utterance. Thus a robust and accurate confidence measure results in better detection of recognition errors, Out of Vocabulary (OOV) words and eventually accuracy of the whole ASR system. Accordingly, the aim of this research has been to obtain several confidence measures from independent knowledge sources to attain the winning combination which leads to more accurate measure. In order to achieve this goal, measures such as ‘acoustic score’, ‘language model score’, ‘decoder score’, ‘speak rate’, ‘gap number’, ‘gap ratio’, ‘uncovered ratio’, ‘turn number’, ‘matched in focus’ and ‘slots matched’ were computed from different modules in a Farsi dialog system which were later fed to Support Vector Machines (SVM) as elements of feature vectors for binary classification of utterances. Thereafter utterance-level continuous annotation was performed through Logistic Regression (LR) based on the same extracted features. The results indicate that combining information from independent knowledge sources leads to 78% correct confidence annotation in continuous scoring and 82% in discrete scoring. Moreover a 5% improvement was achieved comparing to current confidence scoring based on using only one feature.
  9. Keywords:
  10. Speech Recognition ; Confidence Measure ; Out of Vocabulary Words

 Digital Object List

 Bookmark

No TOC