Deteksi Onset Gamelan Bebasis DWPT dan BLSTM

Gamelan consists of various kinds of instruments that have different characteristics. Each has characteristics in terms of the basic frequency, amplitude, signal envelope, and different ways of playing it, resulting in differences in the sustain power of the signal. These characteristics cause the problem of vanishing gradient in the Elman Network model which was used in previous studies in studying the onset detection in the Saron instrument signal which has an average interval of more than 0.6 seconds. This study uses BLSTM (Bidirectional Long Short Term Memory) as a model for training and Wavelet Packet Transformation to design a psychoacoustic critical bandwidth as a model for feature extraction. For the peak picking method, this study uses a fixed threshold method with a value of 0.25. The use of the BLSTM model supported by the Wavelet Packet Transform is expected to overcome the vanishing gradient that exists in a simple RNN architecture. The model was tested based on 3 evaluation parameters, namely precision, recall and F-Measure. Based on the test scenario carried out, the model can overcome the vanishing gradient problem on the Saron instrument which has an average interval between onset of 600 ms. Out of a total of 428 onsets on the Saron instrument, the model successfully detected 426 correctly, with 4 incorrectly detected onsets and 2 undetected onsets. A thorough evaluation for each of the precision, recall, and F1-Measure algorithms obtained 0.975, 0.945 and 0.960.