Sparse time-frequency representations for polyphonic audio based on combined efficient fan-chirp transforms

This work presents a new strategy for obtaining a sparse time-frequency representation of polyphonic audio signals that exhibit continuous pitch changes by combining different instances of the fan-chirp transform, showing for the first time that it is applicable to effectively describe simultaneous audio sources with such characteristics. The method described blends two recent proposals: A fast implementation of the fan-chirp transform based on the structure tensor and a smart combination of time-frequency representations that explores local sparsity. Both methods are further improved in this work: now the former provides better estimates of the transforms' chirp rates by including a filtering stage, and the latter yields smoother and more continuous combinations of the representations. A set of experiments with synthetic and real audio signals illustrates the performance of the method. © 2019 Audio Engineering Society. All rights reserved.