Post-Training-Quantization

[saivinaypsv]

I trained blow attached algorithm and verifed performance with & without POST-TRAINING-QUNATIZATION . HiFi-GAN: High Fidelity Denoising and Dereverberation Based on Speech Deep Features in Adversarial Networks.

Any update on Post-Training-Quantization model for [Bandwidth Extension is All You Need]?

[brentspell]

The paper doesn't have any notes about model quantization, and I have not done any of my own experiments with quantization. Given the similarity of the non-causal wavenet architectures between the two papers, you would probably have similar success with quantization in HiFi-GAN+.