This is a complementary demo for the paper "DRC-NET: DENSELY CONNECTED RECURRENT CONVOLUTIONAL NEURAL NETWORK FOR SPEECH DEREVERBERATION"

• DRC-NET: our proposed model

• DRC-NET_causal: causal version of DRC-NET, by setting the convolution kernel size on time dimension to 1, and using unidirectional LSTM.

• DRC-NET_mono: DRC-NET with single channel microphone input

• DRC-NET_causal_mono: causal DRC-NET with single channel microphone input

• DenseNET: the baseline model[1]

• DenseNET_causal: causal version of DenseNET, by setting the convolution kernel size on time dimension to 1, and using unidirectional LSTM.

• DenseNET_mono: DenseNET with single channel microphone input

• DenseNET_causal_mono: causal DenseNET with single channel microphone input

[1] Z. Q. Wang and D. Wang, “Deep learning based target cancellation for speech dereverberation,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 28, pp. 941–950, 2020