A multi-stage filter for separating speech from background noise

This paper proposes a multi-stage method for filtering speech signals that contain large-amplitude background noise. Many commonly encountered noise sources such as engines, propellers, or turbines produce noise fields that are too spatially incoherent to be effectively filtered using traditional methods. The proposed filtering method uses two filter stages to separate speech from background noise: one to remove the coherent components of the noise, and the other to reduce the incoherent components of the noise. The first stage uses the LMS algorithm. In this stage, strategically placed reference microphones are used to eliminate coherent noise from an array of error microphones. The second stage uses a beamforming algorithm to reduce the remaining incoherent noise. Results are presented that demonstrate the effectiveness of the multi-stage filter.