Digital Music Research Network Seminar

Auditory-model based methods for the pitch analysis of polyphonic music signals
Anssi Klapuri, Institute of Signal Processing, Tampere University of Technology, Tampere, Finland

Wed 15 June, 4:00pm, Room 105

This presentation discusses perceptually motivated methods for the estimation of the fundamental frequencies (F0) of several concurrent sounds in real-world music signals. The main emphasis is laid on practical multiple-F0 estimation methods and not so much on auditory modeling.

I will start out by a brief introduction music transcription and to the subtopics involved in it. I will also discuss the properties of music signals and the basic problems of F0 estimation in them.

Next, I will describe and analyze the processing steps involved in prevailing pitch perception models, analyzing how these affect pitch perception and the practical robustness of pitch estimation in music signals. Especially, I will point out certain advantages that perceptually motivated methods have in the multiple-F0 estimation task.

In the second half of the presentation, I will describe a perceptually motivated but more practically-oriented multiple-F0 estimation method. Typically, these make significant departures from the backgrounding auditory models for practical reasons. In particular, it is necessary to extend the pitch models to the estimation of multiple pitches and, also, modifications are needed to improve the robustness of pitch estimation in the presence of co-occurring sounds and noise. Experimental results and transcription demonstrations with the proposed method are shown.