CONTENT-BASED AUDIO SEARCH

This book is for audio information retrieval§practitioners.§It is about audio content-based search.§Specifically, it is on exploring promising paths for§bridging the semantic gap that currently prevents§wide deployment of audio content-based search§engines. Music search sound engines rely on metadata,§mostly human generated, to manage collections of§audio assets. Even though time-consuming and§error-prone, human labeling is a common practice.§Audio content-based methods, algorithms that§automatically extract description from audio files,§are generally not mature enough to provide the user§friendly representation that users demand when§interacting with audio content. This dissertation has two§parts. In a first part we explore the strengths and§limitation of a pure low-level audio description§technique: audio §fingerprinting. §In the second part, we hypothesize that one of the§problems that hinders the closing the semantic gap is§the lack of intelligence that encodes common sense§knowledge and that such a knowledge base is a primary§step toward bridging §the semantic gap. We present a sound effects§retrieval system which leverages both low-level and§semantic technologies.