A calculating method for the multi-fractal spectrum of speech envelope, frequency wave and pitch contour of speech signal is presented to reflect the change of pitch, energy and frequency more accurately. According to these spectra, 18 novel speech emotion features are proposed, and the clustering performances of these features are analysed. The results of the recognition experiment for seven emotion states demonstrate that the speech emotion features proposed in this paper are effective for the speech emotion recognition, and that they can be used as the complementary features to the acoustics emotion features for the speech emotion recognition.
|