Spectral Complexity Reduction of Music Signals for Cochlear Implant Users based on Subspace Tracking

Spectral complexity reduction can be used to emphasize the leading voice or melody and attenuate the competing accompaniment of music pieces. This method is known to facilitate music perception in cochlear implant (CI) users as spectrally less complex signals are perceived as being more pleasant. In this paper we investigate a method to obtain a reduced-rank approximation for the desired complexity reduction that extends the established projection approximation subspace tracking methods (PAST, CPAST) with an additional sparsity constraint. We evaluate our method with the instrumental SIR and SAR measures as well as an auditory distortion measure (ADR) on a database of 110 classical chamber music pieces. While the resulting signal quality is found to be comparable to existing methods the iterative structure and the reduced computational complexity of our method make it suitable for real-time and low-latency on-line applications.