Pitch-informed solo and accompaniment separation towards its use in music education applications

摘要: We present a system for the automatic separation of solo instruments and music accompaniment in polyphonic recordings. Our approach is based on pitch detection front-end tone-based spectral estimation. assess plausibility using sound technologies to create practice material education context. To better understand quality requirements education, listening test was conducted determine most perceptually relevant signal distortions that need be improved. Results from show tracks pose different should optimized differently. propose evaluate algorithm modifications their effects objective perceptual measures. Finally, we outline possible ways optimizing our suit applications.