State duration modelling in hidden Markov models
Signal Processing
Nonstationary hidden Markov model
Signal Processing
Component-based discriminative classification for hidden Markov models
Pattern Recognition
| Hi-index | 0.00 |
Short-time parameters combined with dynamic classifiers such as Hidden Markov Models have been classically used in pathological voice detection systems, however most of the approaches rely in complex procedures or addition of new parameters that increases the time processing and it does not reflect in a better system performance. This paper presents an approach that improves the standard scheme for Hidden Markov Models based classification oriented to voice pathological identification. This approach uses HMMs to derive discriminative features defined by specific components of individual models, such features span an HMM-induced vector space where a feature based classifier can be trained. Results show that the discussed methodology outperforms significantly the accuracy in a classification system comparing with the standard classification scheme. As a result a high accuracy can be achieved by using a relatively simple procedure to generate an optimal decision boundary. Results are provided using the MEEIVL voice disorder database.