On Vocal Tract Characteristics Of Chinese Whispered Speech And Its Applications In Perceptual Study

Under certain circumstances, people should call in whispering for security or not disturbing others. The research on it can be applied not only to communication but also to medical usage. What's more, the recognition of whispered speech can be used for identity certification by the police and its database is sometimes required by the low-bit coding technique.Without the vibration of vocal colds, we can not abstract the fundamental frequency from whispered speech. Hence, its vocal tract characteristics, including formants, their bandwidths and the gain of vocal tract, become prominent in speech analysis and systhesis. As the trachea with the lung attached to the vocal-tract in whispered speech, its transfer function varies by subjoining zero/pole points. Here, an algorithm for formant estimation of whispered speech based on spectral segmentation is proposed. Experiments show that the proposed method is efficient in formant extraction of whispered speech, especially in low SNR.Chinese is a tonal language, thus it's essential to identify the tune of Chinese vowels. Obviously, tonal analysis is based on the fundamental frequencies which could not be found in the whispered speech. Thus, we proposed another parameter- gain of the vocal tract transfer function to abstract tunes. It is proved to be effectual in tonal analysis of Chinese whispered vowels.Parameter based and non-parameter based methods here are utilized to convert the parameters of vocal tract transfer function from whispered speech to voiced one, which is the basis of speech reconstruction.