Multiclass SVM-Based Isolated-Digit Recognition using a HMM-Guided Segmentation

Automatic Speech Recognition (ASR) is essentially a problem of pattern classification, however, the time dimension of the speech signal has prevented to pose ASR as a simple static classification problem. Support Vector Machine (SVM) classifiers could provide an appropriate solution, since they are very well adapted to high-dimensional classification problems. Nevertheless, the use of SVMs for ASR is by no means straightforward, because SVM classifiers are well developed for binary problems but not so for the multiclass case. In this paper we compare two approaches to implement the multiclass SVM from binary SVMs (1-vs-all and 1-vs-1) for a specific ASR task. We show that the 1-vs-all multiclass SVM clearly outperforms the conventional HMM-based ASR system (the largest improvement, 18.23 %, is achieved for speech corrupted with white noise).