A design method of speech corpus for text-to-speech synthesis taking account of prosody.