低信噪比环境下基于随机共振与谱减法的语音增强

发布时间：2018-04-11 20:30

本文选题：随机共振 + 谱减算法　；参考：《天津工业大学》2017年硕士论文

【摘要】：本文针对谱减算法在低信噪比环境下由于音乐噪声和相位噪声的干扰导致语音增强与检测效果不佳的问题,提出了一种基于双稳随机共振理论与谱减算法的复合型语音增强算法。在对相位噪声的几何分解研究后发现,噪信号相位与纯净相位之间的差值与信号的初始信噪比成负相关,而复合方法利用随机共振在微弱信号处理方向上的独特优势对含噪语音进行预处理,将噪声的能量往低频的语音信号上转移,实现不损伤语音信号的同时提升语音信号的信噪比,从而减小谱减算法中相位噪声的影响。另外,在对预处理过后的语音信号进行谱减时采用自适应增益平均的方式对增益函数做时域上的平滑处理,以此降低谱减算法对语音信号的损伤程度。实验采取Matlab仿真的方式,对含噪语音信号分别通过复合方法语音处理系统与谱减算法语音处理系统,并详细对比分析复合方法与谱减算法输出语音的分段信噪比以及感知语音质量评估得分。经大量实验仿真发现,复合方法与谱减算法对比优势明显,在初始信噪比低于-7dB时复合方法效果尤为显著,可将初始信噪比低至-15dB的含噪语音信号分段信噪比提升至OdB左右,信噪比增益高达15dB。实验结果表明,随机共振与谱减算法结合的复合方法可明显提升低信噪比环境下语音信号的输出分段信噪比及感知语音质量。
[Abstract]:In this paper, the problem of speech enhancement and detection caused by music noise and phase noise in low signal-to-noise ratio (SNR) environment is discussed.A hybrid speech enhancement algorithm based on bistable stochastic resonance theory and spectral subtraction algorithm is proposed.After studying the geometric decomposition of phase noise, it is found that the difference between the phase and pure phase of the noise signal is negatively correlated with the initial signal-to-noise ratio of the signal.The composite method uses the unique advantage of stochastic resonance in the direction of weak signal processing to preprocess noisy speech and transfer the energy of noise to the low frequency speech signal.The signal-to-noise ratio of the speech signal can be improved while the speech signal is not damaged, and the effect of phase noise in the spectral subtraction algorithm can be reduced.In addition, the adaptive gain averaging method is used to smooth the gain function in the time domain in order to reduce the damage degree of the speech signal caused by the spectral subtraction algorithm.The experiment adopts the method of Matlab simulation. The speech processing system of noisy speech signal is composed of speech processing system with compound method and speech processing system with spectral subtraction algorithm.The segmented signal-to-noise ratio (SNR) and perceptual speech quality evaluation score of the speech output from the composite method and the spectral subtraction algorithm are compared and analyzed in detail.A large number of experiments show that the hybrid method has obvious advantages compared with the spectral subtraction algorithm, especially when the initial SNR is less than -7dB, and the segmental SNR of the noisy speech signal with initial SNR as low as -15dB can be increased to about OdB.The SNR gain is up to 15 dB.The experimental results show that the combination of stochastic resonance and spectral subtraction algorithm can significantly improve the output segmented signal-to-noise ratio and perceptual speech quality of speech signal in low SNR environment.
【学位授予单位】：天津工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.35

【相似文献】