Abstract

This study considers a new measure of distortions of speaker speech sounds that is invariant with respect to the gain of speech signal in a communication channel. Properties of the measure are investigated in comparison with its closest analogues. A series of theoretical features has been proved. The new measure is shown to combine advantages of the symmetric Itakura distance in relation to the noise immunity of automatic speech processing, on the one hand, and of the COSH distance in relation to the sensitivity to speech signal distortions, on the other hand. Using the proprietary software, an experiment was set up and conducted. Estimates of the new measure dependence on the signal-to-noise ratio were presented. It has been shown that the logarithmic presentation of this relationship has the pattern close to linear. The obtained results are intended to be used in development of new systems and upgrading of existing systems and technologies for digital signal processing and speech quality analysis under the noise exposure.

中文翻译：

---

用于实时模式下测量语音信号失真的 COSH 距离的标度不变修正

摘要

本研究考虑了一种新的说话者语音失真测量方法，该方法相对于通信信道中的语音信号增益是不变的。与最接近的类似物相比，研究了该度量的特性。已经证明了一系列的理论特征。新措施表明，一方面，对称 Itakura 距离在自动语音处理的抗噪性方面的优势，另一方面，COSH 距离在对语音信号失真的敏感性方面的优势。使用专有软件，设置并进行了实验。提出了对信噪比的新测量依赖性的估计。已经表明，这种关系的对数表示具有接近线性的模式。