Robust Clustering - Based Realtime Vowel Recognition

In the therapy of the hearing impaired one of the key problems is how to deal with the lack of proper auditive feedback which impedes the development of intelligible speech. The effectiveness of the therapy relies heavily on accurate phoneme recognition. Because of the environmental difficulties, simple recognition algorithms may have a weak classification performance, so various techniques such as normalization and classifier combination are applied to raising the overall recognition accuracy. In earlier work we came to realise that the classification accuracy is higher on a database that is manually clustered according to the gender and age of the speakers. This paper examines what happens when we cluster the database into a few groups automatically and then we train separate classifiers for each cluster. The results shows that this two-step method can increase the recognition performance by several percent.