Independent vector extraction under extremely low SNR conditions

This study adopts an independent vector extraction (IVE) algorithm framework to tackle the problem of speech extraction under very low signal-to-noise ratio (SNR) conditions. The reasons why existing IVE algorithms perform poorly in this scenario are analyzed. The narrowing of the objective function’s convergence region makes them prone to getting trapped in local optima, amplitude calibration failure causes distortion of the output signal, and reverberation leads to signal masking. Corresponding improvement strategies are identified and used to optimize an orthogonally constrained independent vector extraction (OGIVE) algorithm and a fast IVE algorithm. Experiments across various scenarios show that the improved fast IVE algorithm outperforms existing methods in extracting weak signals, demonstrating the effectiveness of the proposed improvements.