END-TO-END AUTOMATIC MUSIC TRANSCRIPTION OF POLYPHONIC QANUN AND OUD MUSIC USING DEEP NEURAL NETWORK

Abstract

This paper introduces an automatic music transcription model using Deep Neural Networks (DNNs), focusing on simulating the "trained ear" in music. It advances the field of signal processing and music technology, particularly in multi-instrument transcription involving traditional Turkish instruments, Qanun and Oud. Those instruments have unique timbral characteristics with early decay periods. The study involves generating basic combinations of multi-pitch datasets, training the DNN model on this data, and demonstrating its effectiveness in transcribing two-part compositions with high accuracy and F1 measures. The model's training involves understanding the fundamental characteristics of individual instruments, enabling it to identify and isolate complex patterns in mixed compositions. The primary goal is to empower the model to distinguish and analyze individual musical components, thereby enhancing applications in music production, audio engineering, and education