A Time-Varying Analysis Method for Rapid Transitions in Speech

Abstract

Althongh many time-varying models have been proposed in the speech literature, few, if any, have had success when applied to speech in real applications. This is in part due to the limitations of the models themselves as well as the lack of robustness of their attendant parameter extraction methods. An LPC model based on time-depen dent poles has yielded promising results when applied to synthetic data This paper extends this technique to real speech data. The data are processed pitch synchronously using a simple procedure to identify regions of the data that best fit the model. The maximum-likelihood tech nique, which has been found to be robust in the presence of noise, is used to estimate the parameters. Resulting format estimates for several diphthongs are presented; our algorithm tracks the formants well, both in stable regions and in regions of transition. This ability to track formant variation within analysis intervals is a definite advantage ovei traditional LPC. Results from speech data involving final stop conso nants are presented. Rapid changes, particularly in the first and sec ond formants, in the region immediately prior to the stop have been detected. Such abrupt transistions are often not detected by traditional time-invariant methods. This information about the behavior of formants prior to the stop will be useful as additional information for characterization of stop consonants. © 1991 IEEE