Combining phonetic and corpus-based approaches to remedy disruptive effects in synthetic speech
The project builds upon our previous activities in the field of computer speech processing and aims at basic research into speech synthesis. The key idea in the proposed project is to confront phonetic knowledge of the characteristics of natural speech with the characteristics present in synthetic speech and to exploit the acquired know-how to remedy disruptive effects in synthetic speech and to enhance the performance of corpus-based speech-synthesis techniques. Emphasis will be placed on a thorough analysis of contemporary problems perceived in synthetic speech, identification of relevant phonetic parameters crucial for the quality of synthetic speech and the use of such phonetically justified patterns to drive the latest corpus-based speech synthesis techniques. Special attention will be given to automatic prediction of synthesis errors and to the corresponding signal modification/generation.
The main objective of the project is to remove disruptive effects from synthetic speech by enhancing corpus-based algorithms for generating high-quality synthetic speech while exploiting a thorough understanding of phonetic processes involved in human speech production and perception.
Team of investigators:
The team of investigators consists of two academic workplaces, the team from Department of Cybernetics & NTIS research centre at the Faculty of Applied Sciences, University of West Bohemia (UWB) in Pilsen and the team from the Institute of Phonetics, Faculty of Arts and Philosophy, Charles University in Prague (CU).