Please use this identifier to cite or link to this item:
|Title:||Speech emotion analysis||Authors:||Poh, Esther Ee Chen.||Keywords:||DRNTU::Engineering::Electrical and electronic engineering||Issue Date:||2009||Abstract:||Emotion recognition in speech has become increasingly important mainly due to the prevailing usage of speech in human-computer interaction. However, the studies done in the last century are still not enough to achieve satisfactory accuracies for the speech emotion recognition system. Therefore, this project seeks to investigate the performance of a speech recognition system implemented using the Probabilistic Neural Networks (PNN) under various conditions and parameters. A speaker-gender independent, gender dependent and speaker dependent system were performed using four different feature set namely Pitch and Energy, formants, LPCC and MFCC. In addition, difference in the effect of using a 15-emotion set and a 5-emotion set were also investigated in this project. Experimental results showed that a speaker-dependent system produces higher accuracy as the system is able to account for variability existing between different speakers. It also shows that MFCC outperforms the rest of the speech features in terms of accuracy but lowest performance in terms of speed. Also, a dataset with less emotion classes also proved to improve the accuracies significantly as compared to a larger emotion set due to the reduction in sources of confusion between various emotional states.||URI:||http://hdl.handle.net/10356/18054||Rights:||Nanyang Technological University||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||EEE Student Reports (FYP/IA/PA/PI)|
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.