Interspeech'2005 - Eurospeech

Lisbon, Portugal
September 4-8, 2005

Using Hadamard ECOC in Multi-Class Problems Based on SVM

An-rong Yin, Xiang Xie, Jingming Kuang

Beijing Institute of Technology, China

In this paper, we propose to apply Hadamard Error-Correcting Output Code (Hadamard ECOC) to extend binary classifier for multi-class classification problems. Hadamard ECOC is easy to construct and is suitable for any number of classes. We combine it with binary support vector machine (SVM) to solve the multi-class problem of speaker identification, which takes advantage of error correcting ability of Hadamard ECOC and powerful classification ability of SVM. Compared to the traditional "1-against-rest" method, the experiment result shows that Hadamard ECOC approach has much better and more stable performance for the multi-class problem and is robust on different rules mapping rules between ECOCs and classes.

Full Paper

Bibliographic reference.  Yin, An-rong / Xie, Xiang / Kuang, Jingming (2005): "Using Hadamard ECOC in multi-class problems based on SVM", In INTERSPEECH-2005, 3125-3128.