Research On Speech Enhancement In Airborne Noise Environment

In this paper, the speech enhancement in the airborne noise environment is purposed to reduce or even weaken the single frequency noise and overcome the specific noise, with the help of voice activity detection, adaptive LMS algorithm, speech enhancement algorithm to improve voice quality, and use DSP hardware and software processing platform for noise reduction algorithm transplantation. The main contents are as follows:Firstly, this paper summarizes the theory of speech signal processing, analyzes,discusses and simulates the speech enhancement algorithm, and introduces the subjective and objective evaluation indexes for evaluating the quality of speech and speech distortion.Secondly,aiming at the single frequency (narrowband) noise of the airborne system,two high and low thresholds and a threshold of zero crossings are set to distinguish the single frequency interference signal,and the method of combining the spectral entropy function to detect the speech endpoint, and then use the multi-window spectrum estimation subtraction filtering part of the noise. At the same time,an improved normalized adaptive LMS variable step filteralgorithm is proposed for the on-board tone detection signal. The effect of noise reduction is determined by the minimum error of the input and output of the filter and the convergence speed of the algorithm.and then the processed voice datasare stored in the buffer pool waiting for the output.Thirdly, for the noise spectrum estimation of the low frequency or high frequency components of the airborne environment, the IMCRA algorithm with improved speech enhancement residuals(SER) is used to quickly track and update the rapidly changing nonstationary noise.Through the voice presence probability of the speech frame set the decisionthreshold to distinguish between the silent segment and the speech segment,thereby reducing the noise over the estimation, effectively reduce the delay and deviation;In order to reduce the music noise by improving the LogMMSE algorithm to estimate the prior signal to noise ratio, and smooththe speech power spectrum.The ideal binary masking based on time-frequency division eliminates the threshold portion below the SNR. At last, the enhanced speech signal is obtained, and the MATLAB simulation and data analysis of the improved algorithm are given.Finally, the DSP hardware and software platform are set up to carry out the noise reduction algorithm transplantation. This paper introduces the design of hardware and software based on TMS320VC5416, and analyzes the transplantation idea of spectral subtraction and LMS algorithm.The simulation results show that the speech quality can be improved.