On the Construction of Unit Databanks for Text-to-Speech Systems

摘要

This work deals with one stage in the development of a text-to-speech (TTS) system, which demands a great amount of time and effort, and is strongly related to the resulting speech quality: The determination of the speech-unit databank. For that matter, we present a software tool, the so-called Editor, integrating all major steps in the database determination in a single environment. The whole process includes recording, segmentation, and labeling of speech units to be concatenated in the time domain. The Editor includes a low-cost and precise method for determining the pitch marks, utilizing an auxiliary signal obtained from a contact (throat) microphone. For the phonetic speech labeling, we revise an algorithm for acoustic segmentation, which yields interesting results when proper operation conditions are imposed. The result is a simplified procedure for creating a complete unit database, fully integrated into a single and user-friendly system.