Given the basic nature of the utilized coefficients and the relative simplicity of the applied classification models, the achieved results seem very promising. The tests evaluating the performance of the established classifiers have shown that in all of the considered cases the mean balanced accuracies were greater than 0.900, reaching the maximum of 0.936 (std=0.019) for the C-SVM classifier and the minimum of 0.910 (std=0.017) for the classification using the Decision Tree algorithm.
Logistic Regression, Decision Tree, Random Forest, K-Nearest Neighbors, Linear Discriminant Analysis, Gaussian Naïve Bayes, and C-SVM. The performed experiments involved the computation of the trajectories of fifths for the music pieces from two genre groups - rock/pop and jazz, calculation of their basic coefficients, and using these coefficients as feature variables for various rock/pop - jazz classifiers based on popular machine learning algorithms, i.e. the length of the trajectory) can be utilized as feature variables in music classification algorithms. The results of the conducted experiments indicate that even basic coefficients quantifying the trajectory of fifths (e.g. The study shows that such trajectories provide valuable information concerning the harmonic structure of a given piece of music. In this paper we examine the applicability of the trajectory of fifths as a source of knowledge in automated music classification processes. Alongside with this paper, we publish a novel dataset for extended-vocabulary chord recognition which consists of synthetically generated isolated recordings of various musical instruments. While the joint ACR modeling leads to the best results for isolated instrument recordings, the separate modeling strategy performs best for complex music recordings. Our results show that ACR with an extended chord vocabulary achieves high f-scores of 0.97 for isolated chord recordings and 0.66 for mixed contemporary popular music recordings. We perform a large-scale evaluation using various combinations of training and test sets of different timbre complexity. In our experiments, we compare joint and separate classification of the chord type and chord root pitch class using one or two separate models, respectively. We focus on extending the commonly used major/minor vocabulary (24 classes) to an extended chord vocabulary of seven chord types with a total of 84 classes. In this paper, we build upon a recently proposed deep convolutional neural network architecture for automatic chord recognition (ACR).
0 kommentar(er)
