Automatic chord recognition in digital audio
Date
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
В настоящей работе описывается реализация методов машинного обучения для решения задачи распознавания аккордов в цифровом звуке. Были применены следующие алгоритмы машинного обучения: метод ближайших соседей, метод опорных векторов и наивный байесовский классификатор. Для выбранной аннотированной музыкальной коллекции получены статистические оценки точности, погрешности и распределение результатов по степени точности. Проведен сравнительный анализ результатов работы алгоритмов. В качестве оценок точности были выбраны и реализованы метод кросс-валидации и алгоритм вычисления segment-based Chord Symbol Recall(CSR).
This paper describes an implementation of machine learning methods for the automatic chord recognition in digital audio. The following machine learning algorithms: the Nearest Neighbors Algorithm, Support Vector Machine and Naive Bayesian Classifier were used. Statistics, errors and distribution of the results according of the accuracy for selected annotated music collection were obtained. A comparative analysis of the results of the algorithms is presented. Cross-validation method and algorithm for calculating segment-based Chord Symbol Recall (CSR) was selected and implemented as the accuracy of the estimates.
This paper describes an implementation of machine learning methods for the automatic chord recognition in digital audio. The following machine learning algorithms: the Nearest Neighbors Algorithm, Support Vector Machine and Naive Bayesian Classifier were used. Statistics, errors and distribution of the results according of the accuracy for selected annotated music collection were obtained. A comparative analysis of the results of the algorithms is presented. Cross-validation method and algorithm for calculating segment-based Chord Symbol Recall (CSR) was selected and implemented as the accuracy of the estimates.