Sun Nov 19, 2017 11:45 am
I tried adjusting the fairlight EQ, but it didn't really do much for microphone hiss (Rode VideoMic). I also tried running my files through ffmpeg using highpass and lowpass filters, but the best by far is extracting the audio into Audacity (you need Audacity, ffmpeg dll and liblame installed), doing NR there and then importing the .wav back to Resolve.
It's a bit painful if you need lipsynch. Either the NR or the file formatting means the processed clips run slightly shorter than the original (fraction of a frame per minute). So I had to cut the imported clip into bits and use Alt + left arrow, Alt + right arrow to match the wave form to the original. But it does a good job.
Taking a screenshot of the frequency map in Audacity, then comparing to the frequency map after noise reduction, at first I thought only frequencies over 9000hz were being affected, and I should able to fake that with the EQ in resolve. But looking more closely the frequency map is being changed in subtle ways throughout, so proper NR is more than just low pass/high pass filtering.
I found the defaults for NR good while the file was in Audacity, (-12db, sensitivity 6). But I dropped this down to -8db, sensitivity 4, because on some clips my voice was turning into the King of The Fish People when I finally exported the video. Don't know where that is coming from, so playing it safe and dealing with 80% of hiss rather than 99% seems better to me.