ADAPTIVE NOISE SUPPRESSION IN SPEECH SIGNAL TRANSMISSION.pptVIP

ADAPTIVE NOISE SUPPRESSION IN SPEECH SIGNAL TRANSMISSION Jing Tian, Xiaowen Liu, Zhiming Ni and Mingfeng Li Institute of Acoustics, Chinese Academy of Sciences Beijing 100080, P.R.China 1-INTRODUCTION KEY: reference signal is coherent with the noise in the primary, non-coherent with the desired speech signal. 2-EXPERIMENTAL CONFIGURATION 3-INPUT SIGNAL COHERENCE AND NOISE SUPPRESSION CZ—64 directional microphone for primary signal pick-up The coherence was almost the same in other frequency range, but a difference of about 0.16 was produced at 340 Hz, with the sensitivity difference in the front and back directions is more than 8 dB in 300-2000 Hz 4-IMPROVEMENT ON LMS ALGORITHM SNR IMPROVEMENT SYSTEM PERFORMANCESNRin = -3 dB SYSTEM PERFORMANCESNRin = 3 dB 5-CONLUSIONS The coherence of noise signals and incoherence of speech signals between the primary and reference inputs are equally important for a well designed adaptive noise suppression system applied in speech communications. They could be optimized by proper configuration of the two microphones and the directional pattern of the primary signal pick-up. The modified LMS algorithm with step length variant with both input and error signals performs better, especially when the input SNR is positive. coherence noise suppression performance 1 cm 7 cm SNR8.3 dB SNR12.8 dB Traditional System output Modified The excess mean-square error would increase linearly with the useful signal SNR difference between the primary input and the system output in speech signal intervals, in dB Primary Original speech Modified Traditional Primary Original speech Modified Traditional