Semi-blind source separation using convolutive transfer function for nonlinear acoustic echo cancellation.

Area of Science:

• Signal Processing
• Acoustics
• Machine Learning

Background:

• Nonlinear acoustic echo cancellation (NAEC) is crucial for clear audio communication.
• Existing semi-blind source separation (SBSS) methods for NAEC face limitations in real-time application and stability due to approximations like the multiplicative transfer function (MTF).

Purpose of the Study:

• To propose novel, more effective SBSS methods for NAEC.
• To address the limitations of existing methods, particularly in highly reverberant environments and for real-time processing.

Main Methods:

• Developed two new SBSS methods utilizing auxiliary-function-based independent vector analysis (AuxIVA) and independent low-rank matrix analysis (ILRMA).
• Employed a convolutive transfer function approximation to model long impulse responses with reduced latency.
• Applied carefully regularized optimization schemes tailored to the constrained demixing matrix of NAEC.

Main Results:

• The proposed AuxIVA and ILRMA based SBSS methods demonstrated significantly improved echo cancellation performance compared to existing approaches.
• The use of convolutive transfer function approximation enabled effective modeling of long impulse responses, enhancing suitability for real-time applications.
• The optimized schemes provided a better balance between convergence speed and stability.

Conclusions:

• The proposed SBSS methods based on AuxIVA and ILRMA offer superior performance for NAEC.
• These methods overcome the limitations of MTF approximations, making them suitable for real-time and reverberant conditions.
• The study validates the effectiveness of the proposed techniques for advanced acoustic echo cancellation.