Feature extraction for speech recognition based on orthogonal acoustic-feature planes and LDA

This paper describes an attempt to extract multiple topological structures. hidden in time-spectrum (TS) patterns, by using multiple mappmg operators, and to incorporate the operators into the feature extractor of a speech recognition system. In the previous work, the author proposed a novel feature-extraction method based on MAFP/KLT (MAFP: multiple acoustic-feature planes), in which 3x3 derivative filters were used for mappmg operators, and showed that the method achieved significant improvement in preliminary experiments. In this paper, firstly, the mapping operators are directly extracted in the form of a 3x3 orthogonal basis from a speech database. Next, the operators are evaluated, together with 3x3 simplified operators modeled on the orthogonal basis. Finally, after comparing the experimental results, the author proposes an effective feature-extraction method based on MAFHLDA. in which a Sobel filter is used for mapping operators.