论文部分内容阅读
建立在语音非线性感知特性基础上的谱失真测度,如Mel谱和Bark 谱失真测度等,在语音处理的实际应用中取得了较好的效果.文中提出的余弦镶边临界带滤波谱失真测度也属于这一类,并且综合了二者的优点,所做改进及其特色主要有3 点:一是采用临界带集成原理分配分析滤波器组的中心频率及带宽,使之更加符合耳蜗分析的机理;二是设计了一种新的余弦镶边滤波器代替Mel谱中的三角滤波器,使之对于共振峰的频移不敏感,增强了客观测度在噪声环境中提取共振峰参数的能力;三是具有与Mel谱失真测度相当的计算复杂度,提高了它在实时系统中的可用性
Spectral distortion measures based on non-linear speech perception, such as Mel spectrum and Bark spectral distortion measure, have achieved good results in the practical application of speech processing. The proposed cosine fringes with the critical band filter spectral distortion measure also fall into this category, and the advantages of both, the improvements made and the characteristics of the three main points: First, the use of the principle of critical band with the analysis of the distribution of the filter bank Center frequency and bandwidth to make it more in line with the mechanism of cochlear analysis; the second is to design a new cosine flanger filter instead of the Mel filter in the triangle, so that for the shift of the resonance peak is not sensitive to enhance the objective Measure the ability to extract formant parameters in noisy environments; third, have computational complexity commensurate with the Mel spectral distortion measure and improve its usability in real-time systems