Özet:

Music is still a difficult task today. In this study, general problems were identified from music to emotion recognition, approaches were developed to overcome these problems and increase the success of classification. For this purpose, it was desired to make sense recognition from Turkish music using various machine learning methods and the qualities obtained from different instruments. The method was used by Bayes networks, sequential Minimum Optimization (SMO), Decision Tree (J.48) and Logistics Regression. These methods have been applied to a database created to make emotional recognition and their performance has been measured. This database is a Turkish Emotional Music Database, which consists of 124 music quotes for 30 seconds each. In order to obtain authenticity from musical signals, various tools have been used to provide comprehensive solutions to the problems that we often encounter and encounter during authenticity. These vehicles allow us to obtain a lot of different properties. In addition, the Correlation-based Feature Selection (Correlation-based Feature Selection) has been used to remove unnecessary properties and improve the classification performance. The characteristics obtained from each vehicle are used separately, and the classification process is done along with the methods of machine learning. In the classification phase, a 10-fold cross-verification method has been applied to evaluate and compare the results. In the study, the perception of emotions was performed with 94.35% accuracy by applying the method of selection of characteristics and using the Bayes Networks Classifiers, and the results were better than all of the other classifiers. Finally, all the properties obtained from the vehicles were gathered together and these properties were again selected. After this process, the percentage of emotional recognition obtained by using the Bayes Networks increased by 1.6% to 95.96.

Anahtar Kelimeler: