Research on performance information editing support system for automatic piano - Development of a network model for improved dynamics accuracy-

The automatic piano player, which was previously developed in this laboratory, is attached to the keys and pedals of a grand piano, and enables accurate keystrokes and pedal operation with appropriate control from a computer.To control the device, music data is required, but if music score data is simply input into the device, the performance will be flat, and will not sound like a human being, which is the goal.This is because pianists play with their own intonation when they play.Previous research has developed a system that uses deep learning to predict performance information, but the accuracy of predicting sound volume (Velo) was not good.This research aims to enhance the accuracy of Velo in automatic piano performance.A new deep learning system combining two networks was developed to address limitations in existing methods.