1D CNN-Based Traditional Instrument Classification and Notation Generation in Sinhala for Gataberaya and Flute

Applications for traditional musical instruments from Sri Lanka have not received enough attention, even with the progress in machine learning. This paper presents a novel method for categorizing and transcribing notations for the flute and gataberaya which is an indigenous drum from Sri Lanka using deep learning methods. Based on a manually recorded audio recording dataset with 168 flute recordings and 81 gataberaya recordings with varying lengths and single-noted, 1D Convolutional Neural Networks were used to construct a classification model. The system also generates musical notations from audio recordings using an algorithm and converts Western musical notation to Sinhala musical notations using a prompt with a Large Language Model API call. The methodology includes data augmentation for raw data and feature extraction with Mel Frequency Cepstral Coefficients, and frequency analysis using the Fast Fourier Transform. The classification model achieved 88.71% training accuracy and 91.16% validation accuracy, successfully identifying instrument types, and the notation generation for the flute shows effectiveness by correctly recognizing the notes and the relative octave of the frequency. However, challenges remain for unpitched instruments like the Gataberaya due to the limited dataset. This approach acts as a valuable tool for the preservation of Sri Lankan traditional music.