Modular-Based Classifier for Phoneme Recognition
Abbas Ahmadi, Fakhri Karray, Mohamed S. Kamel · 2006
This paper proposes a modular-based classifier for the problem of phoneme recognition. This is carried out by the use of a two-level classification approach including, high and low levels. We propose a new concept called phoneme family. To obtain phoneme families, we employ k-mean clustering method. A given unknown phoneme is first classified into a phoneme family at high level classification. Then, the exact label of the phoneme is determined at low level classification. We have used a combined framework of statistical and neural network based classifiers. Encouraging results are obtained by applying the proposed method on TIMIT database and its performance is compared against other methods