A new intonation quality evaluation method based on self-supervised learning

Abstract

Intonation evaluation is an important precondition that offers guidance to music practices. This paper present a new intonation quality evaluation method based on self-supervised learning to solve the fuzzy evaluation problem at the critical intonations. Firstly, the effective features of audios are automatically extracted by a self-supervised learning-based deep neural network. Secondly, the intonation evaluation of the single tones and pitch intervals are carried out by combining with the key local features of the audios. Finally, the intonation evaluation method characterized by physical calculations, which simulates and enhances the manual assessment. Experimental results show that the proposed method achieved the accuracy of 93.38% which is the average value of multiple experimental results obtained by randomly assigning audio data, which is much higher than that of the frequency-based intonation evaluation method(37.5%). In addition, this method has been applied in music teaching for the first time and delivers visual evaluation results.