| 跟着科技的成长，语音技巧曾经成为今朝最广泛，最主要的交换手腕之一，然则它却极端轻易遭到净化。噪声经常使语音质质变差，进而作用语音的辨认，分解，编解码等，法语论文范文，终究作用到人们的交换。语音去噪便应运而生，成为语音加强研究中的焦点内容，同样成为语音辨认，编解码处置的前端预处置主要内容。小波去噪因其具有完整的实际基本和辽阔的工程应用空间，成为以后研究旌旗灯号处置的有用办法和热点课题。今朝，随同着各类硬件(如DSP， FPGA， ARM等)的成长和各类技巧的渐进成熟，联合各类语音去噪算法，语音去噪体系逐步适用化。本课题经由过程研究后人的各类经典去噪措施，法语论文题目，先在matlab平台上完成了这些语音去噪体系，用欧式间隔，相干系数，均方根误差目标来评价各自的去噪后果，再斟酌各自的耗时，从而得出较合适硬件完成的办法。然后用各类小波法完成去噪，深刻研究了双正交小波的晋升计划，偏重点研究了晋升小波去噪具有的优势，设计了bior4。4的晋升小波，联合FPGA在DE2开辟板上完成了此晋升小波去噪体系。体系一切的功效模块均是在Alter FPGA上完成的，整合的终究硬件体系调试经由过程，测试成果注解此设计的可行性和准确性，完成了用尽量少的芯片资本，获得较优的处置速度的目标，也注解此硬件体系优于普通的DSP处置器。本文重要的内容：(1)语音和噪声的特点，人耳的感知特征，小波的根本实际及其在去噪中的应用。(2)经典谱减法，各类改良的谱减法语音去噪及matlab平台完成。(3) winner滤波法，自顺应的滤波法完成语音去噪及matlab平台完成。(4)小涉及其晋升小波在语音去噪体系中的应用研究。重点研究了采取分歧阈值，分歧分化层数，分歧处置战略的各类小波的语音去噪后果，并与谱减法，滤波法停止后果比拟，综合评定给出合适硬件及时完成的去噪办法。(5)双正交小波晋升算法去噪体系硬件设计与完成。对各个功效模块一一编程并仿真完成，然后将它们整分解同一的硬件体系，最初对全体体系停止调试，将其成果与matlab平台完成的晋升小波去噪停止比拟剖析，和和传统双正交小波去噪硬件完成成果停止比较剖析。 Abstract: With the growth of science and technology, voice skills has become the most widely used, one of the most important exchange, but it is extremely easy to be purified. Noise often makes the qualitative change of voice quality, thereby affecting the speech recognition, decomposition, coding and decoding process, and eventually affect people's exchange. Speech denoising has emerged as the times require, become the focus of speech enhancement research content, also become the main content of speech recognition, the front end of the codec disposal. Wavelet de noise because of its complete theoretical basis and vast engineering use of space, later to become research signal processing method useful and hot topics. At present, along with all kinds of hardware (such as DSP, FPGA, ARM, etc.) and the growth of various types of gradual maturity, combined with various types of speech denoising algorithm, the gradual application of speech denoising system. This topic through the descendants of the research process of all kinds of classic to denoising method, first in the MATLAB platform, completed the speech denoising system, using Euclidean distance, correlation coefficient, the root mean square error of target to evaluate the respective to the denoising effect and to consider their elapsed time, so as to get the appropriate hardware to complete the way. Then use all kinds of wavelet method to complete denoising, deeply studied the dual orthogonal wavelet promotion plan, partial emphasis on the promotion of wavelet denoising has the advantage of the design of bior4. 4 of the promotion of wavelet, combined with DE2 in the FPGA development board to complete the promotion of wavelet denoising system. All system function module is completed in FPGA, integration of eventually hardware system debugging through the process, the feasibility and the accuracy of the test results of the annotated design, use less as far as possible chip capital, get better processing speed, but also notes the hardware system is better than that of ordinary DSP processor. The main contents of this paper are: (1) the characteristics of speech and noise, the perceptual features of human ear, the basic theory of wavelet and its application in denoising. (2) the classical spectral subtraction, all kinds of improved spectral subtraction speech denoising and MATLAB platform to complete. (3) winner filtering method, the self adaptation of the filter method to complete the speech denoising and MATLAB platform. (4) research on the application of wavelet transform in speech denoising system. Focuses on the discussion take divergence thresholds and divergence layers and different processing strategies of various wavelet speech to the denoising effect, and with the spectral subtraction, filtering method to stop the consequence analogy, comprehensive evaluation given the appropriate hardware the timely completion of the to method of a confusion of voices. (5) the hardware design and completion of the dual orthogonal wavelet promotion algorithm. For every function module one programming and simulation completed, and the whole decomposition with the hardware system, initially on all the system stop debugging, compare its results with the MATLAB platform to complete the promotion wavelet denoising of a comparative analysis, and traditional biorthogonal wavelet to noise hardware complete results of comparative analysis. 目录: |
