Phoneme classification based on supervised manifold learning

This paper proposes an approach for phoneme classification based on supervised manifold learning. It has been shown that speech sounds exist on a low dimensional manifold nonlinearly embedded in high dimensional space and the manifold learning technique can get high phoneme classification accuracy. To improve the performance of phoneme classification, the proposed algorithm calculates the supervised geodesic distance using the minimum distance and the set distance of different class points to enhance the discriminability of low-dimensional embedded data. Experiments show that the proposed algorithm can significantly improve phoneme classification compared to the baseline features.