Abstract: Five approaches of feature extraction: the MEL-frequency Cepstral Coefficient(MFCC), the Linear Predictor Coefficient(LPC), prosodic features, formant frequency and the Zero Crossings with Peak Amplitudes(ZCPA)are described in this paper.These features are applied to emotional speech recognition.According to the recognition results, the weight coefficients of features are obtained by correlation analysis in the three dimensions of PAD emotion model.Simultaneously the recognition results are fused to the PAD emotional space, and the PAD values of the emotional speech are obtained.The PAD values of the emotional speech can be analyzed from the theory of continuous emotion.And the quantitative analysis of emotional speech can reveal the position and relationship of emotional category in emotional space.