Decomposing a polyphonic musical recording to separate instrument tracks has always
been a challenge. Even extracting one of the many instruments in a complex recording
is still unsolved. This paper shows a new way of sound separation on monoaural digital
recordings, that is capable of reliably separating up to 2-3 notes from the recording.
The algorithm uses instrument models based on the analysis of overtone characteristics
of the instrument notes. The paper focuses on the instrument modelling part of the
process.
