Speech Emotion Detection Using Sliding Window Feature Extraction and ANN

Speech Emotion Detection (SED) aims to recognize the classification of each emotion in a continuous speech, and to pinpoint their onset and offset time. Although there is extensive research on emotion recognition, the research about detection is relatively less. In this paper, we propose a novel model for SED using the sliding window method for feature extraction and Artificial Neural Network (ANN) for modeling. As for feature extraction, we firstly introduce the window-level feature which is extracted by emotion recognition utilizing sliding window. Then, using Gaussian distribution to approximate the distribution of emotion contribution from interval-level to window-level in order to calculate the interval-level feature. Finally, ANN is used for model training which uses the fusion of the window-level feature and interval-level feature. We evaluate our model on EMO-DB and experimental results show that the proposed model outperforms baseline significantly.