机载噪声环境下语音增强研究

发布时间：2018-10-05 15:15

【摘要】：本文针对机载噪声环境下的语音增强,以降低乃至消弱单频噪声和克服特定噪声为目的,借助语音端点检测、自适应LMS算法、语音增强算法提高语音质量,并利用DSP软硬件处理平台进行了降噪算法移植,主要内容如下:1、概述了语音信号处理的有关理论,分析、讨论及仿真了有关语音增强算法,介绍了用于评测语音质量和语音失真程度的主客观评价指标。2、针对机载系统的单频(窄带)噪声进行了仿真模拟,设置两个高低门限阈值和一个过零率门限阈值区分出单频干扰信号,与改进谱熵函数结合的方式对语音端点检测,然后利用多窗谱估计谱减法滤除部分噪声,同时对机载的单音检测信号提出一种改进的归一化自适应LMS变步长滤波算法,利用滤波器输入输出端的误差最小以及算法的收敛速度来判定单音降噪的效果,然后将消除单音的语音数据存储在缓冲池等待输出。3、针对机载环境的低频或高频分量的噪声谱估计,利用改进语音增强残差的IMCRA算法,对急剧变化的非平稳噪声实现快速跟踪和更新,通过语音存在概率对语音帧设置判决门限区分无声段和语音段来减少噪声的过估计,有效降低延时和偏差;为了降低音乐噪声借助改进LogMMSE算法估计先验信噪比,对语音功率谱平滑处理,基于时频分割的理想二值掩蔽消除低于信噪比阈值部分,最后获取增强的语音信号,并给出改进算法的MATLAB仿真及数据分析。4、基于TMS320VC5416搭建了 DSP的软硬件平台开展语音降噪算法移植,分析了有关谱减和LMS降噪算法的移植思想,仿真结果表明可改善语音质量。
[Abstract]:Aiming at speech enhancement in airborne noise environment, the purpose of this paper is to reduce or even reduce the single frequency noise and to overcome the specific noise. With the help of speech endpoint detection, adaptive LMS algorithm and speech enhancement algorithm, the speech quality is improved. The noise reduction algorithm is transplanted by using the DSP software and hardware processing platform. The main contents are as follows: 1. The theory, analysis, simulation and simulation of speech signal processing are summarized, and the speech enhancement algorithm is discussed and simulated. This paper introduces the subjective and objective evaluation index .2which is used to evaluate the speech quality and the degree of speech distortion, and simulates the single frequency (narrowband) noise of the airborne system. Two high and low threshold thresholds and a zero crossing threshold are set to distinguish the single frequency interference signal. The speech endpoint is detected by combining with the improved spectral entropy function, and then the partial noise is filtered by multi-window spectral estimation spectral subtraction. At the same time, an improved normalized adaptive LMS variable step size filtering algorithm is proposed for airborne single-tone detection signals. The minimum error at the input and output end of the filter and the convergence speed of the algorithm are used to determine the effect of single-tone noise reduction. Then the single-tone data is stored in the buffer pool waiting for output .3. the improved speech enhancement residual IMCRA algorithm is used to estimate the noise spectrum of the low-frequency or high-frequency components in the airborne environment. The fast tracking and updating of the rapidly changing non-stationary noise is realized, and the decision threshold of speech frame is set to distinguish the silent segment from the speech segment to reduce the over-estimation of the noise, and the delay and deviation are effectively reduced. In order to reduce music noise and estimate prior signal-to-noise ratio (SNR) by using improved LogMMSE algorithm, an ideal binary masking based on time-frequency segmentation is used to eliminate the threshold part of SNR, and then the enhanced speech signal is obtained. The MATLAB simulation and data analysis of the improved algorithm are given. The software and hardware platform of DSP based on TMS320VC5416 is built to transplant the speech noise reduction algorithm. The transplant ideas of spectrum reduction and LMS denoising algorithm are analyzed. The simulation results show that the speech quality can be improved.
【学位授予单位】：华中师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.35

【参考文献】