Turkish Folk Music Modes are Automatically Identified with Artificial Learning Technologies

Yayın Tarihi | 02 March 2024, Saturday

The aim is to explain the concept of maqam, which is one of the fundamental components of Turkish Folk Music, and to investigate how it can be used to automatically identify different maqams. This research aims to create an original compilation in this field, focusing particularly on Turkish folk melodies.

The main aim of this study is to investigate how artificial learning technologies, which have made a significant impact in the world of technology in recent years, can be used to explain the concept of makam, a fundamental component of Turkish Folk Music, and to automatically identify different makams. This research aims to create an original compilation in this field, focusing particularly on Turkish folk tunes. Additionally, the development of technology-based decision support systems and their foundation as educational materials also constitute an important aspect of this study. This study also stands out as one of the few studies where Turkish Folk Music and artificial intelligence techniques come together.

Within the scope of the research, makam classifications were performed on a dataset using different machine learning models. Also, the Turkish Folk Music compilation created for this study and to be shared publicly was tested in this study with the aim of shedding light on other scientific research. In this way, the effect of the relevant models on makam recognition was evaluated. It was noted that the prepared compilation also constitutes an important resource for future research.

Artificial Intelligence Achieves 99.17% Success in Turkish Folk Music!

The research includes eight different machine learning methods (Light Gradient Boosting Machine, eXtreme Gradient Boosting, Naive Bayes, Decision Trees, Support Vector Machines, K-Nearest Neighbor, Random Forest, and Logistic Regression), and a total of 40 different classification processes have been carried out, both with and without data augmentation for each method. The results of the classification processes have been reported with four different parameters: accuracy, precision, sensitivity, and F1 score, resulting in a total of 160 different measurements. The highest accuracy value was achieved with the Light Gradient Boosting Machine classifier using the SMOTE data augmentation technique, with a success rate of 99.17%. The research results indicate that higher performance has been achieved compared to similar studies.

Suggestions for the Development of Mode Recognition Algorithms

In future research, it is suggested to expand and diversify the existing dataset to improve the overall performance of the office recognition algorithms. At the same time, examining the impact of the features in the dataset in more detail can potentially increase the ability to achieve better results. In this context, it is of great importance to evaluate different machine learning algorithms through experiments. In order to achieve the best performance, model selection and hyperparameter optimization need to be addressed in a broader perspective.

Çalışma Linki

Asst. Prof. Dr. Onur SELVI

Faculty of Education - Department of Computer and Instructional Technology Education