Development of Chinese whispered database for speaker verification

A database for speaker verification of Chinese whispered speech is established. It is based on the assumption that whispers are easily affected by the environmental and speakers´ emotional factors. The manuscript for the corpus considers the structure of Chinese syllables, including all the categories of the initials, finals and tones. 8 typical channels are applied to collect the speech, mainly stated as computer-based, communicational mode and record-pen based collection. 7 emotional types, named as calm, angry, amazed, afraid, happy, sad and tired, are designed for the speaker factor analysis. 100 college students, 80 males and 20 females, participated voluntarily in this work. The effectivity of this database is demonstrated in the paper.