Non-negative Matrix Factorization with cepstral distance regularization
Contributors:
Takuya Higuchi, Norihiro Takamune (The University of Tokyo) and Hirokazu Kameoka (The University of Tokyo, NTT Corporation).
Examples of experimental results
Here we demonstrate two examples of unsupervised instrument-wise source separation obtained with the proposed and reference methods using two pieces of music. The audio files are excerpted from the RWC Jazz music database.
Proposed: Joint optimization approach to spectrogram factorization and basis clustering
Reference: Spectrogram factorization followed by basis clustering