A Study On Emotional Information Analse And Process

In order to push the reluctant research in domestic academic circle, I introduced how to analyse and process emotional information of speech in this thesis. Since the relative exploring and experimental method is not seen in the domestic literatures, the methods and data presented in this paper are only preliminary.As an initiative research, we divide the emotion into happiness, anger, surprise and sadness. Then we let several persons who are good in performance and have a good command of mandarin read the texts with four emotional states many times while making the recordings. In order to make the emotional utterances bank effective and be convenient for following processing, we made the effective investigation for every utterance.To account for normal vibrations in speech, nine feature parameters including domain, pitch, energy and formant of speech waveform were calculated. Differences between neutral and emotional speech were found.The recognition research for emotional utterances is the main aspect of the paper. On the basis of main component analysis, we made the templates for four types of emotional speech using large amount or speech data. We presented three methods to recognize the test utterance. The effect shows that the recognition rate are more than 80% which is close to the humane task.