Automatic note generator for Javanese gamelan music accompaniment using deep learning

Javanese gamelan is a traditional form of music from Indonesia with variety styles and patterns. One these patterns the harmony Bonang Barung Penerus instruments. When playing gamelan, resulting can vary based on music’s rhythm or dynamics, which be challenging for novice players unfamiliar rules notation system, only provides melodic notes. Unlike in modern music, where notes are often same all instruments, vital establishing character song. With technological advancements, musical composition generated automatically without human participation, has become trend generation research. This study proposes method to generate accompaniment using bidirectional long-term memory (BiLSTM) network compares it recurrent neural (RNN) (LSTM) models that use numerical represent data, making easier learn variations gamelan. replaces composer completing instruments To evaluate harmonic note distance, dynamic time warping (DTW), cross-correlation techniques were used measure distance between system-generated results composer's creations. In addition, audio features extracted visualize audio. The experimental show produced better accuracy when song, reaching value around 90%, compared 2 (rhythm melody), reached 65-70%. Furthermore, BiLSTM model harmonies more similar original (+93%) than those by LSTM (+92%) RNN (+90%). applied performing music.