Speech Emotion Recognition Based on Speech Segment Using LSTM with Attention Model

摘要

Automatic speech emotion recognition has become popular as it enables natural interaction between human-machine interaction. One modality of recognizing emotion is speech. However, the speech also contains silence that may not relevant to emotion. Two ways to improve performance is by removing silence and/or paying more attention to speech segment while ignoring the silence. In this paper, we propose both, a combination of silence removal and attention model to improve speech emotion recognition performance. The results show that utilizing combination silence removal and attention model outperforms the use of either noise removal only or attention model only.