F0 generation in a text-to-speech system using a database of natural F0 patterns

摘要

This paper describes the new F0 (fundamental frequency) generation model developed for the CPqD (Research and Development Center of TELEBRAS) text-to-speech system. This system produces speech for the Portuguese language spoken in Brazil. The F0 generation model is based on a database of F0 patterns, which are concatenated into new F0 contours. This model is also based on a prosodic parser. The F0 contours produced using this model have been shown to produce high-quality synthetic speech.