Post-Filtering for Residual ECHO Suppression in Under-Modeled Acoustic ECHO Cancellers

This paper investigates the use of post-filtering algorithms for suppressing residual echo in adaptive acoustic echo cancellers, which is an important component of hands-free telephone systems. The goal of post-filtering is to enhance near-end speech corrupted by residual echo due to echo canceller under-modeling. In this paper the use of signal subspace enhancement is investigated as a post-filter structure by employing a subspace decomposition of the echo canceller error signal using the Karhunen-Loeve transform (KLT). In particular, the spectral-domain constrained (SDC) linear estimator is employed to enhance the near-end speech subspace while suppressing the residual echo subspace. Practical and low-complexity implementation considerations are discussed. Simulation results are presented comparing the approach to frequency-domain post-filtering in terms of objective near-end speech distortion and quality measures