Based on concepts which are known to work well with 2-D images, this paper explores the application of a statistical noise reduction method to improving the robustness of speaker identification against background noise. A dynamic-window weighted-rms averaging criterion is used to detect noise and preserve the speech quality. Various forms of noise from the NOISEX-92 database were added to digitised speech to test the enhancement properties of our methods. Experimental results show that the proposed filter leads to improved speaker identification performance, even with certain of our supportedly "clean" speech data sets which were contaminated by high-level background noise. Keywords: Noise Reduction, Speaker Identification.