Optimal Filtering of an Instrument Sound in a Mixed Recording Using Harmonic Model and Score Alignment

In this paper we show a method for optimal filtering (in terms of mean square error) of an instrument in a mixed recording. The filter is based on prior knowledge of the exact pitch played by the instrument and also on the assumption that the instrument is well described within the harmonic model framework. To get the pitch priors, we use alignment of the score information to the real recording. We show how this algorithm can be used to filter a single instrument or voiced singer. We also show how it can be used to substruct an instrument or to balance the volume to several instruments.