Tempo Recognition Of Kendhang Instruments Using Hybrid Feature Extraction

This article is the result of research on Gamelan instruments that examines from a technological perspective what is rarely done nowadays, through kendhang tempo recognition by proposing three classification modeling schemes. The proposed scheme is a new approach to kendhang tempo classification, using kendhang sound converted to image-based features via Mel spectrogram, then features are extracted from the image with Visual Geometry Group (VGG)-19 before incorporating the method K-Nearest Neighbour (K-NN) as a classification method. Based on the experimental results that have been obtained, modeling using the 3rd scheme, namely two-phase feature extraction from the Mel spectrogram image as the first phase and the second phase of VGG-19 with classification using K-NN has an advantage in accuracy (99.6%) of implementing Kendhang tempo recognition correctly and the average achievement of the fastest training processing time was 3.37 seconds compared to the 1st scheme with an accuracy of 94% and an average model training process time of 16.4 seconds and the 2nd scheme with a model accuracy of 98% and the average time to complete the model training process the longest is 6228.6 seconds.