A U-net with Gated Recurrent Unit and Efficient Channel Attention Mechanism for Real-time Speech Enhancement
Sivaramakrishna Yechuri, Sunny Dayal Vanambathina · 2023
Speech enhancement has many real-world applications, like cochlear implants and hearing aids. We propose a Unet with a gated recurrent unit and an efficient channel attention mechanism for single-channel speech enhancement. ECA can be used to implement a novel cross-channel interaction without reducing its dimensionality. Network performance was significantly improved by using an adaptable kernel size in module testing. Additionally, the U-Net architecture uses gated recurrent units (GRU), which yields a causal system suitable for real-world use. GRU is used for learning long-range dependencies. Speakers and noise types are different for training and testing because it is independent of both. Compared to LSTM-based models, UGRU with ECA consistently leads to better objective comprehensibility and perceptual quality.