基于改进小波阈值函数的语音增强算法研究

发布时间：2018-06-18 02:01

本文选题：语音增强 + 小波阈值　；参考：《深圳大学》2017年硕士论文

【摘要】：随着计算机与数字信号处理技术的不断发展,语音信号处理技术也不断完善与成熟,并以视频会议、手机、网络视讯等各种形式广泛应用于现实生活里面。但是各种各样的噪声会干扰语音通讯设备的通讯效果,严重妨碍了人们正常的交流与工作。所以,怎样有效降低噪声带来的问题,提高有用语音信号质量,引起了海内外许多专家研究者们的兴趣。目前也出现了较多的语音增强算法,一些在输入噪声较低情况下增强效果较好,高噪声强度输入下,增强效果不理想。因此,本文在针对各种信噪比输入条件下进行研究。主要做了如下工作或改进:1:分析了小波去噪理论知识,主要深入研究小波阈值函数降噪算法,其目的是保留带噪语音信号中有用语音小波系数,抑制噪声系数。对一些小波阈值函数存在不连续、不同分解层数阈值恒定以及会产生恒定误差等缺点。提出改进的带调整参数连续小波阈值函数,并采用粒子群寻优法寻找改进阈值函数在背景噪声中的最优值,最后将改进的函数与贝叶斯阈值方法相结合,进行小波重构处理后,得到处理后的语音信号系数。将提出的阈值函数与其他已存阈值函数去噪效果对比,仿真实验结果表明在语音输出信噪比、降低有用信号失真及抑制背景噪声等方面有一定提高。2:改进型阈值函数的小波去噪算法在输入噪声强度不高的情况下降噪效果明显,但是在低信噪比下还是残留一些杂声,因此,为进一步提高增强语音去噪算法效果,将带噪语音信号经过改进型阈值函数小波去噪后作为先验信息,再结合卡尔曼滤波算法,得到最终的增强信号。利用MATLAB平台进行实验,其结果表明该结合法能在高噪声输入条件下取得更好的增强效果。3:语音信号采集与显示界面的编写,在PC Windows操作系统上,采用VC++6.0软件编写的MFC界面,通过对USB声卡和Windows所提供的音频Wave类中函数完成对声卡中音频编程,实现语音采集界面。调用Windows所给的绘图API函数,实现波形数据显示功能。
[Abstract]:With the development of computer and digital signal processing technology, voice signal processing technology has been improved and matured, and has been widely used in real life in various forms such as video conference, mobile phone, network video. However, all kinds of noise will interfere with the communication effect of voice communication equipment, which seriously hinders people's normal communication and work. Therefore, how to effectively reduce the problems caused by noise and improve the quality of useful speech signal has attracted the interest of many experts and researchers at home and abroad. At present, there are many speech enhancement algorithms, some of which have better enhancement effect when the input noise is low, but the enhancement effect is not ideal under the high noise intensity input. Therefore, this paper studies on various SNR input conditions. The main work of this paper is as follows: firstly, the wavelet denoising theory is analyzed, and the wavelet threshold function denoising algorithm is studied in depth. The purpose of the algorithm is to retain the useful speech wavelet coefficients and suppress the noise coefficients in noisy speech signals. For some wavelet threshold functions there are some disadvantages such as discontinuity constant threshold of different decomposition layers and constant error. An improved continuous wavelet threshold function with adjusted parameters is proposed, and the particle swarm optimization method is used to find the optimal value of the improved threshold function in background noise. Finally, the improved function is combined with Bayesian threshold method to reconstruct the wavelet. The speech signal coefficients after processing are obtained. The proposed threshold function is compared with that of other existing threshold functions. The simulation results show that the signal-to-noise ratio (SNR) of the speech output is obtained. In some aspects, such as reducing useful signal distortion and suppressing background noise, the wavelet denoising algorithm with improved threshold function can reduce noise obviously when the input noise intensity is not high, but it still has some residual noise under low SNR. Therefore, in order to further improve the effect of enhanced speech denoising algorithm, the noisy speech signal is de-noised by the improved threshold function wavelet as the prior information, and then the final enhanced signal is obtained by combining the Kalman filter algorithm. The results show that the combined method can achieve better enhancement effect under the condition of high noise input. The interface of voice signal acquisition and display is compiled. On PC Windows operating system, MFC interface is written with VC 6.0 software. Through the function of USB sound card and the audio wave class provided by Windows, the audio program of the sound card is completed, and the interface of voice acquisition is realized. The drawing API function given by Windows is called to realize the display function of waveform data.
【学位授予单位】：深圳大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.35

【参考文献】