This item is available under a Creative Commons License for non-commercial use only
Recently, techniques such as shifted 2D non-negative matrix factorisation and shifted 2D non-negative tensor factorisations have been proposed as methods for separating harmonic musical instruments from single and multi-channel mixtures. However, these methods require the use of a Constant Q transform, for which no true inverse exists. This has adverse effects on the quality of the resynthesis of the separated sources. In this paper, a number of different resynthesis methods are investigated in order to determine the best approach to resynthesis
Fitzgerald, D. & Cranitch, M. Resynthesis methods for sound source separation using shifted non-negative factorisation models. Paper given at the Irish Signals and Systems Conference, Derry, Ireland, 2007.