UCSB is digitizing Edison Diamond Discs from the teens and 1920s and
putting them online in DAHR. We are normalizing access files using EBU
R-128 to -16 LUFS. When we do this manually in Wavelab it sounds great.
When we batch process using ffmpeg (double pass), files with certain noise
profiles come out sounding terrible, full of volume pumping. RMS
normalization in ffmpeg works fine. Questions are, does EBU R-128 not work
well with noisy content? Why does the normalizer in Wavelab work so much
better than ffmpeg? And finally, is there any reason not to just use RMS
normalization? Our workflow is automated, so using Wavelab isn't an option.
I can probably share files if anybody want to hear.