|Abstract||Even though there is a rapidly growing corpus of available music recordings, there is still a lack of audio content-based retrieval systems allowing to explore large music collections without manually generated annotations. In this context, the query-by-example paradigm is commonplace: given an audio recording or a fragment of it (used as query or example), the task is to automatically retrieve all documents from a given music collection containing parts or aspects that are similar to it. Here, the notion of similarity used to compare different audio recordings (or fragments) is of crucial importance, and largely depends on the application in mind as well as the user requirements.
In this tutorial, we present and discuss various content-based retrieval tasks based on the query-by-example paradigm. More specifically, we consider audio identification, audio matching, version (or cover song) identification and category-based retrieval. A first goal of this tutorial is to give an overview of the state-of-the-art techniques used for the various tasks. However, a further goal is to introduce a taxonomy that allows for a better understanding of the similarities, and the sometimes subtle differences, between such different retrieval scenarios. In particular, we elaborate on the differences between fragment-level and document-level retrieval, as well as on various match specificity levels found in the music search/match process.