A Dual-Step-U-Net for Crystal-Clear Restoration of Audio Recordings

This work, which is at the forefront of
innovation, uses the powerful capabilities of ConvNet. The
proposed method leverages the power of deep learning by
developing a novel Dual-Step-U-Net architecture to
significantly enhance audio restoration. The proposed
approach analyses the audio's time-frequency
representation and trained on real-world noisy data to
concurrently eliminate common additive disturbances
from vintage analog discs, such as hiss, clicks, and thumps.
Embarking on the forefront of innovation, this thesis
pioneers a revolutionary paradigm in audio restoration and
denoising through the formidable prowess of a fully
convolutional deep neural network.