Multi-band adaptive filtering application on vocal mute

Using a pure audio source such as MP3, CDDA to sing karaoke is widespread now. In order to extract out the background music from a nonstandard recording audio source, a multi-band adaptive filtering architecture is proposed. A psychoacoustics characterization is employed to separate the individual subband features in this application. It has been proved that the effects of the method are better than those of using full band. The probability of divergence of the adaptive tracking system is greatly reduced by using the multi-band method. The quality of music after vocal suppression is subjectively satisfactory, by using the proposed architecture. The information of lyric-synchronization is exploited for weight updating control. It increases the stability of the adaptive tracking convergence, as well as enhancing the music quality.