ACOUSTIC AND PERCEPTUAL INVARIANCE IN SPEECH

点击翻译按钮获取中文摘要

The broad objectives of this proposal are to continue to develop and test a theory of acoustic invariance in speech. To this end, we will extend our research on place of articulation in stop consonants to nasals and fricatives by investigating whether invariant properties can be derived for place of articulation in these two consonant classes. We will also explore the extent to which such properties generalize across these different manners of articulation. For nasals, we will explore the relative changes in spectral energy at high and low frequencies in the vicinity of the nasal release. For fricatives, we will explore the spectral patterns and overall amplitude characteristics of the fricative noise, as well as the patterns of spectral change in the vicinity of the fricative release. In addition, we will investigate a number of phonetic dimensions relating to manner of articulation. We will further elaborate the invariant property proposed for the stop-glide distinction by examining patterns of amplitude change occurring in particular frequency bands in the vicinity of the stop-glide release. Finally, we will attempt to determine if invariant properties can be derived for the nasal-stop contrast and the stop-affricate-fricative contrast. For the nasal-stop distinction, we will explore the phonetic feature [nasal] by investigating the spectral properties of the nasal murmur and the presence of nasalization in the vowel formants. For the stop-affricate-fricative contrast, we will focus on patterns of amplitude change in the early portions of the consonant release. Our general research strategy includes performing detailed acoustic analyses of natural speech tokens and conducting perception experiments using synthetic and computer-edited natural speech. The goals of the acoustic analyses are to identify those invariant properties relating to particular phonetic features, to devise algorithms for quantifying these properties, and to test the effectiveness of these measures by applying them to a new set of natural speech data. Perception experiments will be conducted to help us focus on potentially critical portions of the acoustic signal where invariant properties are likely to reside, and to explore the perceptual consequences of these properties for the identification of particular phonetic dimensions.