A Classification Approach to Automatic Music Transcription

The emergence of large music collections has elicited a great deal of research on methods for content-based music retrieval and organization. In our recent research, we address the long-standing problem of automatic music transcription, i.e. the process of converting an audio recording to a musical score. Whereas previous approaches attempt to model acoustics such as the periodicity of harmonic components, we pursue the insight that explicit prior knowledge is not strictly necessary by examining a classification approach to music transcription. In this talk, we will provide a brief overview of previous methods and outline our efforts for the transcription of melody and polyphonic piano music.