Karaoke system with spatial acoustics estimation for vocal or instrumental removal

This paper presents a new approach and prototype system for karaoke vocal or instrument removal. The algorithm estimates the singer's location and spatial transfer function from a stereo recording, then mutes components in the recording according to the estimated orientation. At the current stage, estimation is obtained by analyzing solo segments in the recording. This algorithm turns to be robust against reverb and spatial acoustics. To compensate low frequency loss that usually appears after vocal removal, the original recording is low-pass filtered around 120Hz and mixed back with the processed mono sound to generate a stereo karaoke track.