A statistical analysis of power-level-difference-based dual-channel post-filter estimator

• We study the power-level-difference-based dual-channel post-filter estimator under a stochastic model.
• We verify the theoretical results by using Monte Carlo simulation results.
• We propose four schemes to improve the traditional PLD-DCPF estimator.

By deriving the explicit expression of the probability density function (p.d.f.), this paper presents a statistical analysis of the power-level-difference-based dual-channel post-filter (PLD-DCPF) estimator. The derivation is based on the joint p.d.f. of the auto-spectra of a two-dimensional stationary Gaussian process with mean zero, where the theoretical expression is verified by numerical simulations. Using this theoretical p.d.f. expression, this paper studies the impacts of the correlative parameters on the amount of noise reduction and speech distortion. According to both the theoretical analysis results and the simulation results, four schemes are proposed to improve the performance of the traditional PLD-DCPF estimator.