Information Retrieval and Recommendation Using Emotion from Speech Signals

摘要

In this paper we describe a system of retrieving information from artwork based on textual cues, descriptive to relative art pieces, made available through the metadata itself. Large datasets of artwork can easily be mined by using alternative queries and search methodologies. In the most common search methodology a text-based query using a keyboard is performed. We are proposing a method for searching, finding and recommending digital media content based on pre-set metadata text queries organized in two categories, then mapped to speech sentiment cues extracted from the emotion layer of speech alone. We also account for the difference in sentiment expression for male and female speakers and further suggest that this differentiation may improve system performance.