Characterisation of Emotions in Speech

7017 Words29 Pages
Characterization of emotions in speech using excitation source information This paper explores the excitation source features of speech production mechanism for characterizing the emotions present in speech. The excitation source signal is obtained from speech signal using linear prediction (LP) analysis and it is also known to be LP residual. For the voiced speech the excitation signal looks like a periodic sequence of impulses with varying strengths and for unvoiced speech the excitation signal appears like random noise. Speech signal has high signal to noise ratio in the glottal closure (GC) region, hence it is important to process the speech or excitation source signal around instants of glottal closure to perform various speech tasks. In the excitation source signal, the significant excitation (impulse like) will take place at the instants of glottal closure during the production of voiced speech. These instants of significant excitation are also known as epochs. The following excitation source features are proposed in this paper for characterizing the emotions. : (1) The sequence of LP residual samples, to indicate the higher order relations present in the excitation source signal, (2) The phase signal corresponding to the LP residual, (3) Parameters of epochs ( instantaneous pitch, strength of excitation, sharpness of excitation, slope of the strength of excitation ) (4) Parameters of the glottal pulse (length of opening phase, length of closing phase, length of complete closure, slope of opening, slope of closing, speed coefficient= length of opening/ length of closing) and (5) Dynamics of the epochs and glottal pulse parameters at the syllable and utterance levels. For developing the emotion specific models auto associative neural network models are explored. Berlin emotion speech corpus is considered for this study. In these studies six basic emotions

More about Characterisation of Emotions in Speech

Open Document