Tone Recognition of Chinese Whispered Speech

This paper concentrates on tone recognition of Chinese whispered speech. Without fundamental frequencies, four-tone classification for Chinese has been assumed extremely difficult. Former studies suppose that it is the amplitude of whispered speech can be utilized for tone recognition. However, the amplitude of speech can be changed on purpose, especially in whispers. Then assumption has been made that formant track can indicate tones, while subsequent handling is still on expectation. In this paper we proposed an algorithm on this topic. It can be mainly divided into three steps. First, endpoint detection and Initial/Final segmentation of whispered speech. Second, curve fitting for gain track of whispered vowels. Last, ,four-tone classification. Tests are carried on Chinese whispered speech and the proposed algorithm is proved to be efficient.