Research Seminars

Integrating Context and Content to Search for Music via Natural Language Queries

Peter Knees
Department of Computational Perception, Johannes Kepler University Linz

16:00 Wednesday 9 April 2008

Abstract

The talk will cover methods to automatically build a search engine for large-scale music collections that can be queried through natural language text input. While most existing approaches depend on explicit manual annotation and meta-data assigned to the individual audio pieces, the presented system derives descriptions automatically by building upon methods from Web Retrieval and Music Information Retrieval. Using the information found in the ID3 tags of a collection of mp3 files, relevant Web pages are retrieved via Google queries and used to characterise the music pieces. The additional use of acoustic similarity allows for describing audio pieces with no associated Web information and improves retrieval performance.

In the course of the talk, I will discuss different approaches to build the music piece representations and to query the system. Furthermore, I will demonstrate the ability of the approach to incorporate relevance feedback to improve search results, i.e. to accommodate to the user's preferences based on the user's choices.