異種楽器音の分離に関する研究

概要

論文の詳細を見る
The study of transcription for computer has been in progress in recent years. Many systems of transcribing performed music have been proposed. But most of these systems treat of monophonic input, so we can hardly find studies that treat of polyphonic input. This paper proposes a system to extract pitches (fundamental frequencies) and names of instrument from polyphonic input, which consist of two musical instruments of different class. The main feature of this system is as follows ;(1) This system is possible to apply to tones of short duration in music by using MEM (Maximum Entropy Method), which has high frequency resolution with short data, for the method of pitch extraction (2) In source discrimination, this system becomes concise by using the property of transient portion. For this system, we performed a simulation for two kind of polyphonic input. One is a different class of instruments. The other is a same class of instruments. As the results of simulations ; the discrimination rate for a different class of instruments showed 92.6% for 300 times. In conclusion, this system is valid to extract pitches and names of instruments for different class of musical instruments.
前橋工科大学の論文
2000-03-25