Single-channel blind separation using pseudo-stereo mixture and complex 2-D histogram

Area of Science:

• Signal Processing
• Audio Engineering
• Machine Learning

Background:

• Single-channel blind source separation (SCBSS) is challenging due to limited input data.
• Existing SCBSS methods often require iterative optimization or prior source information.
• Developing efficient and accurate SCBSS algorithms remains an active research area.

Purpose of the Study:

• To introduce a novel single-channel blind source separation (SCBSS) algorithm.
• To demonstrate the algorithm's advantages, including pseudo-stereo signal creation, independence from initialization and prior knowledge, and avoidance of iterative optimization.
• To validate the algorithm's performance and computational efficiency against existing methods.

Main Methods:

• Modeling source signals using an autoregressive process for characteristic estimation.
• Constructing separation masks directly from the single-channel mixture.
• Formulating a pseudo-stereo mixture by time-shifting and weighting the original signal.
• Estimating artificial mixing system parameters using a weighted complex 2-D histogram.
• Deriving the separability of the proposed mixture model and identifying conditions for maximum likelihood-based mask construction.

Main Results:

• The proposed SCBSS algorithm achieves superior performance compared to existing methods.
• The algorithm demonstrates significant computational speed advantages.
• Experimental results on synthetic and real-audio sources confirm the effectiveness of the novel approach.

Conclusions:

• The developed SCBSS algorithm provides a computationally efficient and effective solution for separating audio sources from a single channel.
• The method successfully mimics stereo signal concepts and operates without requiring iterative processes or prior information.
• The approach offers a promising advancement in the field of single-channel blind source separation.