Note Detection and Multiple Fundamental Frequency Estimation in Piano Recordings

Abstract

Automatic music transcription (AMT) is a difficult signal processing problem, which has, in the past decade or two, begun to receive proper treatment. An overview of the problem with a focus on the nature of music signals is given, and two significant AMT challenges are addressed in detail—note onset detection and multiple fundamental frequency estimation. Recent work on these problems is summarized, and an algorithm considering both challenges in the context of piano audio transcription is proposed. A portion of the algorithm concerning multiple fundamental frequency estimation is, to the knowledge of this author, unique. The algorithm is tested, and results are shown for a recording of Bach’s BWV 847 fugue.