Audio Upscaling with Deep Learning
Frequency fused residual convolutional network to resolve high frequency data
With the continued growth of videoconferencing it is imperative that sound quality is maintained regardless of microphone quality. This work focuses on upscaling low resolution audio using deep learning to solve this problem. We have successfully implemented a novel frequency residual neural network that increases the frequency range of low sample-rate audio clips as well as tested the effectiveness of generative adversarial networks and other low quality data forms on the upscaling process.
