A Decoupled Filtered-x Lms Algorithm for Listening-room Compensation

In hands-free scenarios the desired speech signal picked up by the microphone is corrupted by various disturbances such as additive noise, acoustic echoes, and room reverberation. Especially the cancelation of room reverberation still remains a challenging task. For time-variant acoustic environments adaptive filters with appropriate learning algorithms based on the well-known least-mean-squares (LMS) algorithm can be used. Examples known from the field of active noise control (ANC) are the filtered-X LMS (FxLMS) or the modified filtered-X LMS (mFxLMS). In this contribution a decoupled version of the mFxLMS with a faster convergence speed will be introduced. Furthermore, an overclocking of the filter update can be applied which allows for even faster convergence at the cost of additional computational load. The new algorithm is evaluated under realistic environments including ambient noise and estimation errors of the room impulse response (RIR).