Pruning a Model Object

[mert-salahi]

Hi all,

First of all - thanks for all the work that went into this. This is a huge help for someone starting out in this research area.

I'm looking to train a speech enhancement model and ultimately run inference on a resource-limited edge device. So model pruning and quantization is of particular interest to me. Has any one experimented with doing this within the asteroid framework? I know BaseModel inherits from a vanilla PyTorch nn class, so I imagine it should be possible to use the PyTorch pruning API.

Another question: has anyone implemented the CRUSE model in the asteroid framework? This looks like a good model to try given my resource constraints.

Thanks!