Abstract
We address the problem of blind source separation from a single channel audio source using a statistical model of the sources. We modify the Bark Scale aligned Wavelet Packet Decomposition, to acquire approximate-shiftability property. We allow oversampling in some decomposition nodes to equalize sampling rate in all terminal nodes. Statistical models are trained from samples of each source separately. The separation is performed using these models. The proposed psycho-acoustically motivated non-uniform filterbank structure reduces signal space dimension and simplifies training procedure of the statistical model. In our experiments we show that the proposed algorithm performs better when compared to a competing algorithm. We study the effect that different wavelet families have on the performance of the proposed signal analysis in the single-channel source separation task.
Original language | English |
---|---|
Pages (from-to) | 339-350 |
Number of pages | 12 |
Journal | Journal of Signal Processing Systems |
Volume | 65 |
Issue number | 3 |
DOIs | |
State | Published - Dec 2011 |
Keywords
- Audio source separation
- BS-WPD
- CSR-BS-WPD
- CWT
- GMM
- Monaural source separation
All Science Journal Classification (ASJC) codes
- Control and Systems Engineering
- Theoretical Computer Science
- Signal Processing
- Information Systems
- Modelling and Simulation
- Hardware and Architecture