Speech emotion recognition using recurrent neural networks with directional self-attention

作者：

Highlights：

• Directional self-attention is introduced to bi-directional long-short term memory.

• Autocorrelation of speech frames is explored with Self-Attention mechanism.

• The frames with rich affective information are paid more attention.

• The emotion recognition architecture automatically annotates the weights of frames.

摘要

•Directional self-attention is introduced to bi-directional long-short term memory.•Autocorrelation of speech frames is explored with Self-Attention mechanism.•The frames with rich affective information are paid more attention.•The emotion recognition architecture automatically annotates the weights of frames.

论文关键词：Speech emotion recognition,Bi-directional long-short term memory with directional self-attention,Self-attention,Autocorrelation

论文评审过程：Received 18 September 2019, Revised 1 February 2021, Accepted 3 February 2021, Available online 12 February 2021, Version of Record 3 March 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2021.114683