7-444 - Real-time Automatic Piano Music Transcription System

Dasaem Jeong

Abstract: Research in automatic music transcription (AMT) showed significant improvement thanks to advances in deep learning. However, most of the research was designed for an offline scenario, where the input audio recording is provided from beginning to end. In this paper, we present an online piano AMT system with visualization using a web browser and MIDI export which works on CPU in real-time. We employed a model with auto-regressive LSTM and multi-note-state, which is adapted for an online scenario without losing its accuracy.