The thesis addresses the use of machine learning methods for recognizing musical chords from audio signals. It focuses on the development and implementation of Convolutional Neural Networks (CNN) and Support Vector Machines (SVM) for analyzing and classifying chords.
The problem of chord recognition is presented in the context of audio signal processing and feature extraction, such as spectrograms and chromagrams.
The thesis provides a detailed presentation of the CNN architectures used for pattern recognition in spectrograms and the SVM methods adapted to work with audio features. The effectiveness of these approaches is compared based on experiments with different configurations and parameters.
The results show that CNN models provide a high level of accuracy in chord recognition, while SVMs offer faster execution with slightly lower accuracy. The analysis of the results reveals the advantages and disadvantages of each approach and provides insight into possibilities for further development aimed at improving speed and accuracy.
Key conclusions include the confirmation that the combination of machine learning and knowledge from music theory is effective in chord recognition, and that further progress in this field requires ongoing research and the integration of advanced machine learning methods.
|
