机构地区: 深圳大学
出 处: 《计算机工程与应用》 2007年第23期80-82,共3页
摘 要: 在谱减去噪过程中噪音的功率谱估计一般根据经验而定,对此提出了一种改进方法,利用含噪语音的短时能零积和基本谱减法,得到语音起止点和噪音功率谱估计,有利于在不同语音阶段对含噪语音进行谱减去噪。最后利用无音阶段噪音特点对去噪之后的残留噪音进行残差处理以彻底去除噪音。仿真实验表明该方法比传统单一的谱减去噪方法效果理想。 In the course of de-noising using spectrum subtraction,power spectra of noise is generally decided via traditional experience.A new approach is proposed to change the traditional approach.Speech start-end points and power spectra of noise are firstly achieved through general spectral subtraction methods and little-time energy-zero-product.Then further denoising through spectral subtraction methods is more effective during the different respective speech segment.Residual noise reduction is proposed based on the characteristic of noise in the non-speech activity.When this method is applied to denoise speech corrupted by additive white Gaussian noise,simulation experiment results illustrate that this method is more effective than the traditional spectral subtraction methods.