Answers are grounded in this paper and cite their sources.
A Dilated Time Frequency Attention Autoencoder (DTFAAEC) model for the real-time speech enhancement is proposed which consists of a fully convolutional neural networks with time frequency attention (TFA). TFA blocks have been followed by the layers in the decoder and encoder. TFA mechanism is designed to learn important information in time, channel and frequency in Convolutional Neural Networks (CNN). At different resolutions the context aggregation is helped by the dilated convolutions. To avoid the information flow from future frames, casual convolutions are used, therefore we will make the network which is applicable for the real-time applications. For upsampling we use the sub-pixel convolutional layers in the decoder. The experimental results shows better performance than the existing techniques in terms of quality scores and objective intelligibility.
No takes yet. Share an insight, caveat, or question.
Vanambathina et al. (2023) studied this question.