Five types of emotions were analyzed and compared in previous paper. The proposed method was completely simulated on PC; the results demonstrate that significant advances have been achieved in this area. The wavelet transform and improved HMM make our speech emotion recognition system robust. The combination of wavelet transforms and HMM as a solid model that can reduce error rates.
Table 1. The result using wavelet transform and HMM
Emotion A F J S P
Anger 85 6 1 8 0
Fear 14 78 6 1 1
Joy 0 0 89 1 10
Sadness 1 3 4 92 0
Surprise 0 0 16 4 79
Total 101 87 116 106 90
Table 2. The result using wavelet transforms and improved HMM
Emotion A F J S P
Anger 90 4 1 4 1
Fear 11 83 3 1 2
Joy 0 0 95 2 3
Sadness 0 1 2 97 0
Surprise 0 2 7 3 88
Total 101 90 108 107 94
When compare table 1& 2 there is certain improve in efficiency but many shortages are lying the selecting of the features, so to find more efficient features and to do further analysis and experiment in a wide field.
Keeping problems discussed in previous chapter following will be my goals, which will be achieved during my thesis:
• The emotion is to be detected from the input speech signal the whole signal processing revolves around the speech signal for the extraction and selection of speech features correspond to emotions.
• In the previous work there is only single wavelet is used so to improve efficiency, the crossbreed decomposition takes place.
• The next is generating a database for training and testing of extracted speech features followed by the last stage of emotion detection by the classifier section using pattern recognition algorithms.
• Extraction and selection of speech feature: - The extraction of speech feature involves potential audio segmentation followed by acoustic preprocessing like filtering to form their meaningful units. The purpose of the audio segmentation is to segment a speech signal into units that are representative for emotions
• Database for training and testing: - A good database is as important as the desired result. There are different database created by speech processing community used in research work.
• Detect emotions: - Stored database are classifier to detect the emotions by comparing the vectors from the trained data and test data vector.