Some of the NR reduction algos, such as the Spectral (possibly the best one), work on 'bins', splitting the audio into 'bins' and then selecting which ones have audio in them or not etc.
Currently (v1.0) we run the audio path at a full 44kHz sample rate, which is 'wasteful', as we are normally only interested in the audio part, which is probably upto 3.5kHz max.
If we decimate the audio path by 4, that takes us down to ~11kHz, so an audio bandwidth of ~5.5kHz. That will still leave us with plenty of 'space' for normal audio, and even an OK bandwidth for AM SWL. It will hopefully allow the NR systems to 'focus' more finely on the actual audio, and thus hopefully do an even better job of sorting out the noise from the useful sounds.
Some of the NR reduction algos, such as the Spectral (possibly the best one), work on 'bins', splitting the audio into 'bins' and then selecting which ones have audio in them or not etc. Currently (v1.0) we run the audio path at a full 44kHz sample rate, which is 'wasteful', as we are normally only interested in the audio part, which is probably upto 3.5kHz max.
If we decimate the audio path by 4, that takes us down to ~11kHz, so an audio bandwidth of ~5.5kHz. That will still leave us with plenty of 'space' for normal audio, and even an OK bandwidth for AM SWL. It will hopefully allow the NR systems to 'focus' more finely on the actual audio, and thus hopefully do an even better job of sorting out the noise from the useful sounds.
Try it out...