Calibration Method to Measure Accurate Bow Force for Real Violin Performances

In this paper, we present a procedure to predict bow pressing force in a violin from data acquired in real recordings. We focus on the calibration procedure that must be robust to the bow tension changes in long recordings and fast enough to not disturb the recording session. Because of this last limitation, the calibration method here proposed does not exhaustively cover all the possible bow conditions that potentially may appear in the recording. We propose the use of Support Vector Regression to predict all these missing scenarios and compute the predicted force. On the other hand, bow tension variations in long recordings produce decalibrations on the acquisition system. After analyzing their behavior, we propose a solution to compensate this effect based on post processing and a specific behavior of the performer at the beginning and the end of each phrase.