Gamelan Orchestra Transcription Using Neural Network
Dewi Nurdiyah, Yoyon Kusnendar Suprapto, Eko Mulyanto Yuniarno · 2020
Orchestra transcription is a complex task in AMT (Automatic Music Transcription) because it contains a mixture of many notations from various instrument sources. In this paper we propose a reliable method for the transcription of notation of the balungan group gamelan orchestra consisting of demung, saron and peking instruments. Orchestra recordings did not initially have a ground truth label. Therefore, the method we propose is divided into two parts. First, constructs ground truth by recognizing each instrument in the orchestra with bandpass filter. Then, detect the pitch notation and its duration using adaptive threshold and frequency threshold. Second, orchestra signals in the time-frequency domain is trained together with the ground truth that has been built using a neural network. Our proposed method has been evaluated and produce significant testing performance with an average accuracy is 0.9658.