Optimization of Tone Recognition via Applying Linear Discriminant Analysis in Feature Extraction

F0 is an important tone features in the state-of-art tone recognition system. Traditionally, difference of F0 (F0), subsection slope and intercept, and subsection mean F0 and mean F0, are used to improve the recognition accuracy. In fact, all these features can be expressed as the linear transform of F0. The problem is to find the best coefficients for the transform. Linear discriminant analysis (LDA) is a good methodology in finding an optimal linear feature subspace. This paper introduces the LDA methodology to optimize the tone feature extraction in tone recognition. The critical steps of LDA are deduced and the advantage of LDA is theoretically argued. Experimental results on isolative syllable database confirm that LDA-based features perform much better than other features.