Spectral Shape Factors as Acoustic Invariants for Speech Perception (Computer and Information Science)

本研究旨在探讨一种基于短时语音频谱形状因子的语音信息感知模型。频谱形状因子被定义为语音频谱的余弦基向量展开的低阶分量。本研究的主要目的是确定频谱形状因子是语音的基本信息承载参数，还是仅仅是一种方便的语音频谱编码方法。对于元音和顿音，基于频谱形状因素的感知模型将与主要基于频谱峰的感知模型直接对比。在这两种情况下，将进行实验以确定语音信号中的声学不变性可以用频谱形状因子表示的程度。本研究不仅有助于对语音信息编码过程的基本理解，而且对听力障碍语音自动识别的研究和感觉替代领域的工作也有潜在的帮助。

This research is aimed at investigating a perceptual model for representing information in speech based on spectral shape factors of short-time speech spectra. Spectral shape factors are defined as the low-ordered components of a cosine basis vector expansion of speech spectra. The primary objective of the proposed research is to determine whether spectral shape factors are fundamental information-bearing parameters for speech, or whether they are merely a convenient method for encoding speech spectra. For both vowels and stop consonants, a perceptual model which is based on spectral shape factors will be directly contrasted with a perceptual model which is based primarily on the peaks in the spectrum. In both cases experiments will be conducted to determine the extent to which coustic invariance in the speech signal can be represented by spectral shape factors. This research will not only contribute to a fundamental understanding of the information-encoding process in speech, but also has the potential for aiding automatic speech recognition research and work in the area of sensory substitution for the hearing impaired.