Real-time ML audio noise suppression on Raspberry Pi Pico 2

the DeepFilterNet 2 paper has benchmarks which include the RNN model used in the article. TLDR is that DFN2 does a better job.

I suspect they went with the RNN model because the Pico 2 could run it with the 10ms max latency they wanted.

DFN2 is fast, but I'm not sure it's lightweight enough for something like the Pico 2.

That samsung link crashes my browser every time I open it; quite impressive really :)