Swarthmore College Department of Computer Science

Derek Tingle -- CCSCE Best Student Poster

Derek Tingle's poster Reverse Engineering the Music Genome: Automatically Annotating Music with Tags won the best student poster award at this weekend's Consortium for Computing Sciences in Colleges (CCSC) Eastern Conference at Villanova University.

Congratulations, Derek!! dt picture

Reverse Engineering the Music Genome: Automatically Annotating Music with Tags

Abstract

The popularization of digital music has led to the creation of new ways for listeners to discover music. "Music Discovery Engines", such as Pandora, are web-based systems that recommend new music by finding songs based on the user's interests. These interests can be represented as semantic text-based tags (e.g., "southern rap influences", "prominent use of the banjo", "energetic beats"). The most straightforward way to annotate music with tags is to use human experts. That is, a trained musicologist will listen to a song and select the tags that describe it. This process is time consuming and costly. We present a system that automatically annotates music with tags based on an analysis of the audio signal. To do this, we train a probabilistic classifier that learns a relationship between audio features and tags. We show that we can automatically annotate a large corpus of 10,000 songs using a vocabulary of 400 tags from Pandora's Music Genome Project.

Faculty Advisor: Doug Turnbull