랜덤 마스킹 및 Seq2Seq 모델 기반 악기 디지털 인터페이스 데이터의 향상된 평가 방법

Abstract

In this study, an enhanced evaluation method for MIDI data based on random masking and Seq2Seq model was proposed. This method is intended to evaluate neither the creativity of musical works nor the aesthetics of the MIDI data composed by AI models. The model is used to analyze the features of MIDI data to evaluate their quality. Specifically, in the proposed method, the random mask processor should be used to mask MIDI data and train a Seq2Seq model to analyze the knowledge of basic MIDI data theory and analyze MIDI data features to evaluate the generated MIDI data quality automatically without general descriptive parameters. The proposed method could be used to overcome the limitations of subjective evaluation MIDI data.