A voice-to-MIDI system for singing melodies with lyrics
Naoki Itou, Kazushi Nishimoto · 2007
In this paper, we propose a robust Voice-to-MIDI (V to M) system with which a user can input MIDI sequence data by naturally singing melodies with lyrics. A Voice-to-MIDI system translates singing voices into digital musical data, i.e., MIDI sequence data. Therefore, with such a system, users can input melodies intuitively, which releases them from manual translating memorized melodies into chromatic pitches. However, the quality of translation of ordinary Voice-to-MIDI systems is insufficient. One of the most significant problems is the poor accuracy of the segmentation of notes. We solve this problem by employing "rhythmic tapping" concurrently with singing. We examined the proposed method by the accuracy of the numbers of segmented notes and their pitches. As a result, we confirmed that our system outperformed ordinary Voice-to-MIDI systems. Thus, this system satisfies both of easy and intuitive composition of MIDI sequence data and high accuracy of translation of sung data into MIDI sequence data.