Wavelet Model for Vocalic Speech Coarticulation

摘要

A known aspect of human speech is that a vowel produced in isolation (forexample, ee) is acoustically different from a production of the same vowel in the company of two consonants (deed). This phenomenon natural to the speech of any language, is known as consonant-vowel-consonant coarticulation. The effect of coarticulation results when a speech segment (d) dynamically influences the articulation of an adjacent segment (ee within deed). A recent development in the theory of wavelet signal processing is wavelet system characterization. In wavelet system theory, the wavelet transform is used to describe the time-frequency behavior of a transmission channel, by virtue of its ability to describe the time-frequency content of the system's input and output signals. The present research proposes a wavelet-system model for speech coarticulation: wherein, the system is the process of transformation from a control speech state (input) to an effected speech state (output). Specifically, a vowel produced in isolation is transformed into an effected version of the same vowel produced in consonant-vowel-consonant, via the coarticulation channel. Quantitatively, the channel is determined by the wavelet transform of the effected vowel's signal, using the control vowel's signal as the mother wavelet. A practical experiment is conducted to evaluate the coarticulation channel using samples of real speech. The results show that the model is capable of depicting coarticulation effects associated with certain vowel-consonant combinations. They suggest that elements of the vowel's acoustic composition are continuously present, in a modified form throughout the consonant-vowel transition. For other phonetic combinations. however, the model does not respond to instances of segmental transition in a character.