Text-to-speech synthesis using a natural voice source