Fourth European Conference on Speech Communication and Technology

Madrid, Spain
September 18-21, 1995

A Method for Fully Automatic Analysis and Modelling of Voice Source Characteristics

V. Darsinos, D. Galanis, George Kokkinakis

Wire Communications Laboratory, University of Patras, Greece

In this paper, a fully automatic pitch synchronous method to derive the voice source characteristics starting from the speech signal itself, and fitting them to a parametrized model, is presented. It does not require additional signals for the closed-phase alignment, like the ECG, and uses a fitting procedure that guarantees a good convergence. Further- more the method seems to work reliably even for the closed phase where the analysis interval becomes short. For the automatic finding of the glottal closure instants the method uses a modified version of a previous epoch-detection method. This modification consists of the application of linear detectors to improve performance and reliability mainly in the high-frequency energy speech signals. The method does not require user interaction during the analysis execution. In addition, the required computational load permits its integration in a personal computer environment.

Full Paper

Bibliographic reference.  Darsinos, V. / Galanis, D. / Kokkinakis, George (1995): "A method for fully automatic analysis and modelling of voice source characteristics", In EUROSPEECH-1995, 413-416.