Development of an intelligent model for musical key estimation using machine learning techniques

Every piece of music is characterized by its key, melody, harmony, metre, and rhythm. Musical information retrieval tasks like transcription, chord estimation, and harmony analysis require musical key data as the fundamental comprehension for their implementation. Even though several investigations were carried out by researchers aimed at developing an optimum key profile for a given melody, the possibilities of finding the key using machine learning techniques have been least explored. In this paper, we present a novel approach to determine the musical key of a given song. The proposed model features a simple architecture for learning and classification. It was tested with four distinct machine learning algorithms namely K-nearest neighbor (KNN), Naive Bayes (NB), Discriminant Analysis (DA), and Support Vector Machine (SVM). In addition, a dataset of different genres of music has been compiled, for our experiments. The Pitch Class Profile (PCP) distribution of our dataset has been compared with renowned datasets and it showed similar distribution with the others. We optimized our model with the best classifier from all the four machine learning techniques we used. Out of the four machine learning algorithms used in our model, the SVM gave an accuracy value of 91.49% with the highest precision and recall values. The KNN approach showed an accuracy of 89.76% followed by Naive Bayes and the Discriminant Analysis classifiers with an accuracy of 87.11% and 86.77% respectively. Also, the error rates of these different approaches ranged from 8.51% to 13.23%. These results show that the proposed model with SVM algorithm has a considerably higher accuracy value, and in comparison with recent publications, it is evident that our model can play a pivotal role in the efficient determination of keys since it brings together information related to musical theory and supervised learning techniques for classification.

Automated summary

This paper presents a novel approach for determining the musical key of a given song using machine learning algorithms. The proposed model achieved high accuracy and demonstrated its potential for efficient key determination.