search

Le-Xiaohuai-speech / DPCRN_DNS3

Implementation of paper "DPCRN: Dual-Path Convolution Recurrent Network for Single Channel Speech Enhancement"
188 stars 41 forks source link

Any plans about a smaller and stronger the pretrained models？ #7

Open luuuuuupo opened 2 years ago

luuuuuupo commented 2 years ago

First, thanks for the open-source implementation. I saw a smaller dpcrn model discussed in open questions and readme. Do you have any plans for the releasing schedule? Thanks a lot!

Le-Xiaohuai-speech commented 2 years ago

I have already trained a model with only 1.1G MACCs which can run real-time inference on a Raspberry Pi-4b. I will try to compress the model further in the next few months. There are a lot of methods to choose, and I will spend some time researching and implementing and write a paper, and then open source the implementation.

wendongj commented 2 years ago

I have read an article in zhihu about some advanced denoising model, and the author said the DPCRN have about 7.45Gflops, it really make a great progress with only 1.1G MACCs, hope to see your implementation in the later, since it is really important to have a small complexity when imploy it in the mobile device.