Dimensionality reduction-based spoken emotion recognition