24-09-2016, 03:22 AM
(23-09-2016, 05:41 PM)Olli H Wrote: I’m far from the purist. History of rock’n’roll ...
Can't remember the exact values, but I think LU meters are basically K-filtered RMS with different longer (slower) integration times. I believe there's some gating applied too. I use RME's TotalMix and DIGICheck for monitoring and metering, in a same way as you use your setup. TotalMix is basically just a separate DSP monitoring mixer used in RME gear, which allows me to put the DAW and computer operating system audio into separate busses, and then level match these. It also provides a bunch of other cool features like true mono and true side summing into center speaker, lt/rt fold down of surround audio without change in levels etc, and these can then be controlled from a desktop remote controller totally independent from the software. DIGICheck then provides all kinds of different metering 'modules' for audio passing through TotalMix from simple digital peak to highly customizable spectrum analyzer and correlation meters, every imaginable and fully customizable level meters to pure digital bitstream and noise statistics etc. Very useful and time saving tools.
I use Spotify (Premium) for reference tracks in music. I wrote a UNIX Bash script for batch EBUR128, ReplayGain and RMS analysing files using ffmpeg and sox binaries. In Spotify I generated a radio playlist including music from all sorts of different genres and eras, and recorded the bitstream for a couple of hours. Then I sliced the songs into separate files and batch analyzed them. Common factor for every song that had been normalized downward was ReplayGain value +6, which is basically 'same' as -9 LUFS Short Term, indicating that this is the limit. I took down the computer audio bus in TotalMix by 12 dBs, resulting in situation where music (the beefy parts) from Spotify 'always' plays around -23 LUFS and peaks at -21 LUFS Short Term. I can then A/B Spotify and DAW from the TotalMix remote controller level matched. Later I did the analysis also for YouTube and iTunes.
I also use the pseudo mastering mentality, as we are uploading encoded files, it is very reasonable to cut away as much of the unnecessary headroom as possible before encoding.
About integrated program loudness, and why I think it's not a good way for normalizing music. Some highly personal opinions coming up, and I am bracing myself for a certain type of storm, but you mentioned something about listening reasonable arguments, so I'll try to conjure up something. As you stated, it would be pointless trying to work towards some integrated loudness based on metering, as the measuring would be impossible. The integrated loudness normalization works very well on the broadcasting platform, where there simply aren't enough human resources to check and correct the material. The actual normalization is mostly done by automated offline processing, based on the assumption that the production is done by professionals and the sound is ok, but the level might need final adjustments. This is in fact quite pragmatic and 'crude' part of the post production, and is done by automation on music streaming services too. When I used integrated normalization with music, I came to the conclusion that the loudness 'experience' between different styles and genres could have huge, and for me, unwanted differences, even if they measure the same integrated loudness. This is of course understandable because of the different styles in the composition, arrangement, overall tone of the instrumentation and so on. In another words, the meanest rock song might be basically anchored to the -23 through the song with just a slight level build in the final chorus, while the forte fortissimo in the grande finale of a classical piece might hit -18 sounding hugely more powerful, as might the most intense part of lead solo of an acoustic jazz trio. I think the audible loudness of this climax point despite the genre should be the common reference point of the loudness 'experience', so that no matter what this level is or the overall dynamics of the piece, it would be the same for all material. This would of course mean that generally louder music like rock and EDM would be relatively louder in integrated measurements, which, in my opinion, would be actually kind of natural, because these genres have a loud nature, and should be experienced louder. This is the way loudness normalization is done for music in most streaming services, probably because most people do find this the most pleasant and natural way of listening music, and I find it kind of odd that there are some people promoting strongly against this. Maybe it's the mastering engineers who have based their income on producing those -5 LUFS hitting masters, or the investors who bought shares in classical or jazz record labels, and want the streaming services to switch to high headroom integrated normalizing, as the contemporary pop would start to sound weak and bad in comparison.
And then there is the human ingenuity. It's actually a small miracle, how the most popular contemporary pop sound has fairly quickly evolved the almost perfectly opposite characteristics of which the LU measuring emphasizes; the bass is ultra low drone tone, so it passes the K-curve 'unnoticed', the beats are composed of super short transienty samples, which pass through the slower integration time windows, vocals are done in rapping style, and longer notes are edited choppy for the same reason. All the legato elements are filtered to a very narrow bandwidth. This type of sound then slips through the LU measurement algorithm and sounds very impressive when it explodes out from the charts and playlists. And most certainly the young producers have developed this style by intuition to fight the loudness normalization they probably don't even know exists, and just try to achieve sound loud as possible as it is perceived better. And the best thing is these normalization algorithms were partly developed to battle the loudness war in the first place.
