A METHOD FOR ASSESSING THE EFFECTIVENESS OF SIMILARITY MEASURE FOR AUDIO CLUSTERING USING AREA UNDER THE CURVE (AUC).

Abstract

Audio fingerprint analysis is a widely used method for identifying specific pieces of audios, and can be effective at analyzing specific audio tracks. However, this analysis is only effective in detecting exact duplicate content that matches another piece of audio. A method for finding similarity between audio pieces is reviewed, and an evaluation method to find how effectiveness of that algorithm is proposed. The evaluation method uses Area Under the Curve (AUC) which is calculated by using similarity ranking.