Chinese Musical Instrument Source Separation Demo

CISM is an innovative deep learning model specifically designed for separating traditional Chinese musical instruments. While existing research has primarily focused on Western instruments and vocals, CISM addresses the unique challenges posed by the distinctive timbres and rich playing techniques of Chinese traditional instruments.

Key Features:

• Frequency-Band Division Strategy: Based on the discovery that Chinese instruments exhibit concentrated spectral energy in specific frequency bands, CISM employs a novel frequency-band division approach that significantly reduces computational costs while maintaining high-quality separation.
• Instrument-Specific Processing: The model analyzes and processes four traditional instruments (Guzheng, Dizi, Pipa, and Xiao) using customized frequency band divisions tailored to each instrument's acoustic characteristics.

Through extensive experimental evaluation, CISM has demonstrated both computational efficiency and excellent separation quality, marking a significant advancement in the field of Chinese musical instrument source separation.