Impact of future frames on the performance of the DNN and LSTM in terms of STOI improvement (in %). The input contains 11 past frames, a current frame and {0, 1, 2, 5, 8, 11} future frames. The models are evaluated with six unseen speakers and the unseen babble noise. (a) Performance of the DNN and LSTM at −5 dB SNR. (b) Performance of the DNN and LSTM at −2 dB SNR.