Fig. 2

Overview of the time-domain adaptive attention network for single-channel speech separation. To better extract the local information and global information in speech features, two different attention networks are used in this model: (1) the CBAM is introduced in local attention networks to focus on subtle details of the speech signals (frame-level features). (2) In the global attention networks, the transformer based on self-attention mechanism is used to explore the global associations of the speech contexts (utterance-level features). (3) In the ablation experiment, we replaced the local attention module and/or global attention module with the RNN block