Speaker factor analysis of whispered speech from global spectral features

Speaker factor analysis of whispered speech from global spectral features is proposed. According to the perceptual experiments, the Arousal-Valance factor is imported to determine the speaker's state. The spectral parameters from the Sinusoidal Model and Auditory Model, in addition to the Short-term Spectral Features, are abstracted and tracked. The global statistics from all of the variables mentioned above are calculated to identify the speaker's sentiment of whispered speech. The experimental results indicate that the accuracy of this system reaches to 90%. This classification method and speaker factor description scheme offer an effective path to state analysis of whispered speaker.