The performance of deep neural networks has been shown to be very effective in multiple arenas. As their use becomes more widespread, increased focus is being placed on how implementable they are on a wider variety of devices. These devices often include those with low enough processing power that the operation of the neural network has a significant impact on storage and computation resources. Here we investigate the impact of pruning weights from a deep generative convolutional auto-encoder with skip connections. The chosen model is trained in a Generative Adversarial Network (GAN) setting for audio post-processing of a low bitrate audio coder. We evaluate performance using a combination of objective scores and listening tests.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.