Note: Your custom schedule will not be saved unless you create a new account or login to an existing account.
| Paper: | SP-P13.15 |
| Session: | General Topics in Robust Speech Recognition |
| Time: | Thursday, May 20, 13:00 - 15:00 |
| Presentation: |
Poster (ICASSP 2003 Presentation) |
| Topic: |
Speech Processing: Confidence Measures/Rejection |
| Title: |
CONFIDENCE MEASURES FOR KEYWORD SPOTTING USING SUPORT VECTOR MACHINES |
| Authors: |
Yassine Benayed; LORIA | | |
| | Dominique Fohr; LORIA | | |
| | Jean Paul Haton; LORIA | | |
| | Gerard Chollet; ENST CNRS-LTCI | | |
| Abstract: |
Support Vector machines (SVM) is a new and very promisingclassification technique developed from the theory of Structural RiskMinimisation. In this paper, we propose an alternativeout-of-vocabulary word detection method relying on confidence measuresand support vector machines. Confidence measures are computed fromphone level information provided by a Hidden Markov Model (HMM) basedspeech recognizer. We use three kinds of average techniques asarithmetic, geometric and harmonic averages to compute a confidencemeasure for each word. The acceptance/rejection decision of a word isbased on the confidence feature vector which is processed by a SVMclassifier. The performance of the proposed SVM classifier iscompared with methods based on the averaging of confidence measures. |
| |
| Back | |