Abstract
A new method for vowel phoneme signal recognition is proposed based on principal component analysis. First, the subspace of each vowel signal is estimated by principal component analysis on sevaral vowel signals. Further, the vowel signal to be recognized is divided into frames, and each frames is separately projected into each of the obtained subspaces. In theory, the representation error obtained by projecting the signal into its corresponding subspace is smallest, based on which the vowel phoneme signals can be accurately recognized, The effectiveness of the proposed phoneme recognition method on PCA is proved by experiments.