We adapt the multi-channel approach by (Islam et al., 2019) and build a multi-channel BLSTM classifier. We uses multiple channels BLSTM with a feedforward attention mechanism, which outperforms the state-of-the-art emotion recognition algorithm.
check our paper: https://asistdl.onlinelibrary.wiley.com/doi/pdfdirect/10.1002/pra2.255
run MC-BLSTM and MCNN respectively.