Multiple Pitch Estimation for an Automatic Music Transcription System Using a Support Vector Machine

M. Fujieda, T. Murakami, and Y. Ishida (Japan)


Multiple Pitch Estimation, Automatic Music Transcription, and Support Vector Machine.


In this paper, we propose a new approach to multiple pitch estimation for an automatic music transcription system using a support vector machine (SVM). The SVM is often utilized for identifying musical instruments of sounds. However, there are very few conventional methods to use the SVM for the pitch extraction. The multiple pitch estimation by the SVM is robust against difference of timbre, and does not involve the particular processes to detect an octave. Additionally, our approach effectively uses decision functions of the SVM. As the result, the proposed method requires only single tones as training samples. In experimental results using MIDI, we achieved about 98.71 % accuracy.

Important Links:

Go Back