【摘要】：語音信號不僅具有蘊含信息量最多的特性,而且還是語音信號處理領域的重要組成部分�，F(xiàn)實生活中語音信號是非平穩(wěn)的時變隨機信號,都會受到噪聲的污染,因此必須進行語音信號去噪處理。傅立葉變換、短時傅立葉變換、小波變換對非平穩(wěn)隨機信號處理的效果不好,Hilbert-Huang變換具有的多分辨率和高度自適應的特性使得它非常適合處理非平穩(wěn)非線性的時變隨機信號,本論文提出了基于Hilbert-Huang的語音信號去噪算法研究。首先,介紹了語音和噪聲的基本特性,傅立葉變換、短時傅立葉變換、小波變換的基本理論,常見的語音信號去噪方法的去噪原理和常用的語音信號的語音質量的評價標準。其次,詳細闡述了Hilbert-Huang變換的基本理論和算法實現(xiàn)過程,分析了Hilbert-Huang變換的EMD分解和Hilbert變換的解析過程,并仿真實現(xiàn)了三種不同的信號的Hilbert-Huang變換。再次,根據(jù)語音信號的短時平穩(wěn)性、三次樣條插值法擬合信號曲線的平均包絡的過程中產生的欠沖和過沖的現(xiàn)象和Hilbert-Huang變換過程中噪聲和有用信號在IMF分量中能量分布的不同選取篩選分界點問題,提出了改進的Hilbert-Huang語音信號去噪算法。最后,利用matlab進行仿真對比,分別采用小波變換、Hilbert-Huang變換和本文改進的Hilbert-Huang變換對加噪后語音信號進行處理,仿真結果顯示:本文改進的Hilbert-Huang算法具有更好的去噪效果,去噪后的語音信號不僅有較好的時頻波形,而且有較高的信噪比。
[Abstract]:Speech signal not only contains the most information, but also is an important part of speech signal processing. In real life, speech signal is non-stationary time-varying random signal, which will be polluted by noise. Therefore, speech signal denoising must be carried out. The effect of Fourier transform, short time Fourier transform and wavelet transform on non-stationary random signal processing is not good. Because of its multi-resolution and highly adaptive properties, Hilbert-Huang transform is very suitable to deal with non-stationary and nonlinear time-varying random signals. In this paper, a speech signal denoising algorithm based on Hilbert-Huang is proposed. Firstly, the basic characteristics of speech and noise, the basic theory of Fourier transform, short-time Fourier transform, wavelet transform, the de-noising principle of common speech signal de-noising methods and the evaluation standard of speech quality of common speech signal are introduced. Secondly, the basic theory and algorithm of Hilbert-Huang transform are described in detail, the EMD decomposition of Hilbert-Huang transform and the analytical process of Hilbert transform are analyzed, and the Hilbert-Huang transform of three different signals is simulated. Again, based on the short-term stationarity of the voice signal, In the process of fitting the mean envelope of the signal curve by cubic spline interpolation, the undershoot and overshoot phenomena produced in the process of fitting the mean envelope of the signal curve and the different selection and selection of the boundary points for the energy distribution of the noise and useful signals in the IMF component during the Hilbert-Huang transformation are also discussed. An improved Hilbert-Huang speech signal denoising algorithm is proposed. Finally, matlab is used to simulate and compare, and wavelet transform, Hilbert-Huang transform and the improved Hilbert-Huang transform are used to process the noisy speech signal, and the wavelet transform, the Hilbert-Huang transform and the improved Hilbert-Huang transform are used to process the noisy speech signal. The simulation results show that the improved Hilbert-Huang algorithm has better denoising effect. The de-noised speech signal not only has better time-frequency waveform, but also has a higher signal-to-noise ratio.
【學位授予單位】：長春理工大學
【學位級別】：碩士
【學位授予年份】：2016
【分類號】：TN912.3
，

本文編號：2472239