The Acoustic-to-Articulatory Mapping of Voiced and Fricated Speech
Abstract
Acoustic-to-articulatory mapping is the estimation of a time-varying vocal-tract shape from an acoustic waveform. While most research in acoustic-to-articulatory mapping considers only purely voiced speech, this dissertation investigates the problem for speech that includes fricatives. Aspects of fricative production and perception challenge many of the assumptions and techniques used in existing acoustic-to-articulatory mapping algorithms. This work investigates these issues and extends existing techniques for the acoustic-to-articulatory mapping of purely voiced speech to unvoiced and voiced fricatives in isolation and in continuous speech. Linked-codebooks are used to examine the acoustic-to-articulatory mapping of voiced and unvoiced static fricatives. Acoustic-to-articulatory mapping performance is evaluated by analyzing articulatory estimation error for a number of synthetic fricatives and phonetic class clustering for a collection of real fricatives. Scatter plots of acoustic-to-articulatory mapping results on unvoiced fricatives demonstrate good phonetic class clustering and inter-class separability. For equivalent performance on voiced fricatives, the acoustic features had to be modified to deemphasize frequencies below 1 kHz. Linked-codebook lookup, along with dynamic programming, is used to perform acoustic-to-articulatory mapping of continuous, purely voiced speech. Direct application of the algorithm to speech containing fricatives suggests that purely voiced acoustic-to-articulatory mapping provides contextual information that can improve fricative acoustic-to-articulatory mapping. A five step procedure is developed for the dynamic acoustic-to-articulatory mapping of continuous, voiced speech containing intervocalic fricatives. A collection of vowel-fricative-vowel tokens is used for development and testing. In most cases, the estimated articulatory
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1997
- Accession Number
- ADA334781
Entities
People
- Edward L. Riegelsberger
Organizations
- Air Force Research Laboratory