Unsupervised Low-Rank Representations for Speech Emotion Recognition

Abstract

We examine the use of linear and non-linear dimensionality reduction\nalgorithms for extracting low-rank feature representations for speech emotion\nrecognition. Two feature sets are used, one based on low-level descriptors and\ntheir aggregations (IS10) and one modeling recurrence dynamics of speech (RQA),\nas well as their fusion. We report speech emotion recognition (SER) results for\nlearned representations on two databases using different classification\nmethods. Classification with low-dimensional representations yields performance\nimprovement in a variety of settings. This indicates that dimensionality\nreduction is an effective way to combat the curse of dimensionality for SER.\nVisualization of features in two dimensions provides insight into\ndiscriminatory abilities of reduced feature sets.\n