Setting proper AC3 dialog normalization

I'm currently working on a project to covert a looooong movie from PAL to NTSC. It's split across 3 DVD.

My typical method for determining dialog normalization in a 'from scratch' project involves doing an RMS normalization scan with Sony SoundForge.

For PAL2NTSC projects, I will usually examine the original audio track, and use the same normalization they do (figuring they have commercial tools that can calculate this better).

In this movie, the PAL audio is -31db (on all 3 discs). When I tried that number for my test disc, I found myself occasionally turning the volume down to about -32/-33db in the louder portions.

Typically, with my set up, -28 +/- 2db is my normal listening level and NEVER turn it below -30 to -31db.

So I tried -27 (figuring it seems to be 4 db louder than my normal levels). It's largely ok there, but I end up listening around -26db and there still seems to be a good deal of swing between loud and soft dialog, let alone loud music and effects. I ended up reaching for the remote quite a bit. (A good indicator of the wrong normalization.)

I'm not completely clear about testing the RMS with SoundForge in a 5.1 situation. One guide says test the whole Center channel, another says highlight and test just a section with dialog. If I don't have an original AC3 track to examine, it's usually 2 channels and I do well scanning the full film (both channels).

Trying different methods I get wildly different numbers between discs, channels, segements tested, etc. (Between -27 and -35.)

Any suggestions what to do here? I'm thinking of just using -31db and turning down the volume. I suppose as inconsistent as the audio is on this disc, being commercial produced their number is probably more correct.

I just don't know.

I've been through all this before.

There is no accurate and reliable method for hobbyists like us to measure LAeq. While RMS measurements are a useful indication, they don't take into account subjective or "perceived" loudness.

In fact, it seems that the AC3 tracks on most commercial DVDs have been encoded with dialnorm left at the default of -27dB, so I suspect that some professionals don't fully understand it either.

I briefly played with the Replay Gain measurement in LAME (think it gives a value as an offset from -20db instead of -31db) and initially thought I was onto something, but after further testing found that results were not so reliable.

You want to measure only clean dialogue spoken at normal conversation level, so yes on a 5.1 track that's just from the centre channel. Furthermore, it is the long term average level you are trying to estimate - if there are pauses in a conversation, that can skew the readings to the low side. You could try, as I did, editing a section of dialogue to cut out all the gaps between words and carrying out measurements on that. Great fun to hear the results, but you may still find that different sections of dialogue in the film have varying levels.

The LFL Pwnage DVD I did has 3 audio tracks all with different relative levels. In the end, dialnorm for each track was set using trial-and-error. Although the actual dialogue ended up being louder on the tracks with a lower dynamic range, the final levels are set so that the louder sections of these tracks are still at a reasonably high level.

Hmm. Well that clarifies things, but doesn't get me much closer to solving the problem.

Edit: Testing disc 2 (which is the one I was currently working with) and trimming the center channel to be largely dialog... I end up around -31db +/- 3db... which doesn't help if the loud passages are too loud even at -29db.

I think I'll stick with -28.  It should get my normal listening to around -27/-28 and keep the loudest from exceeding -31... I think.