This paper reports on the development of an English speech synthesis system for a Japanese amyotropic lateral sclerosis (ALS) patient as a part of the project of developing a bilingual communication aid for this patient. The patient had a tracheotomy three years ago and anticipating a possibility of losing his phonatory function. His English speech database for Festival, a free speech synthesis system, was generated from his reading of a US diphone list. There were two problems with the recording. The first was the noise the artificial ventilator made and the second was the difficulty in pronouncing English. For the former issue, noise reduction method was proposed and was found effective. The latter issue was not only a problem with this speaker but also a general problem with non-native speakers of English since English vowels are very difficult for Japanese to pronounce. Although the speaker's English database was successfully built by Festvox and the voice was recognized as his voice, the utterance was unintelligible. We, therefore, proposed reconstructing the patient's database by partially mixing the English native speaker's database. The results showed that the proposed approach can be a promising approach.
展开▼
