DeepFilterNet input to Whisper.cpp

[StuartIanNaylor]

The results from https://github.com/Rikorose/DeepFilterNet are excellent and should dramatically increase whisper recogintion in the presence of pretty high noise levels.

https://github.com/Rikorose/DeepFilterNet also contains a binary and a LADSPA plugin but wondering if it can be tweaked and integrated to give even less load.
Its approx x3 realtime on a single core of my Rk3588 Arm based A76/A55 SBC.

[StuartIanNaylor]

You know I am not sure this is such a great idea the more I play with whisper as its debatable if DeepFilterNet actually helps, maybe on the smaller models with attenuation turned down low.
Playing with DeepFilterNet has got me playing with the larger models of Medium & Large and they work better without DeepFilterNet on crazy SNR levels.
I never tested Whisper with noisy audio before but compared to other ASR with noisy audio its results are jaw dropping and it doesn't even seem to be mentioned all that much.

Does anyone have any ideas how OPenAi are processing audio with either Medium.en or Large as its crazy good and in a totally different league but haven't a clue how?

The artefacts of Deepfilter have a huge effect on Whisper whilst it seems impervious to surrounding noise without ?