Audio analysis - Incorrect True peak?

[macg]

Hi,

I'm running some DCP tests to prepare for an upcoming project, and I've noticed some inconsistencies in the audio analysis in DCP-o-matic.

First, I had a situation where I got different True Peak values from the same audio file. I don't remember the exact sequence of steps that led to the different results, so I can't reproduce it reliably, but I was analyzing the same file and got different True Peak readings at different points.

I also noticed that the colour of the numbers in the audio analysis is not always consistent. For example, a value can appear in yellow, but if I lower the gain and then raise it back to the original value, the exact same value can sometimes appear in black instead.

I also checked the files in DaVinci Resolve. The True Peak readings are significantly different (approximately 1 dB of difference), although the Integrated Loudness (LUFS) values are quite similar. I've seen this behaviour with two different audio files.

I have upgraded to the newest version of DCP-o-matic, but I'm having the same results.
I'm using the app for windows for the first time. I have used on a Mac before and don't remember having these issues.

So my questions are:

1. Could I be doing something wrong in the way I'm analyzing the audio?
2. Is there any known issue that could cause the audio analysis, particularly True Peak, to produce inconsistent or incorrect results?
3. Would you recommend changing the gain in DCP-o-matic to get the True Peak to -3 dB, and can I rely on this measurement?

I appreciate any help! Thank you!

[carl]

Hi,

I doubt you are doing anything wrong. I can't remember hearing about any problems in this area.

The colour of the numbers sounds like a straightforward bug. I'll try to reproduce that.

I'll also try to compare DoM to Resolve. I am quite clueless with Resolve, however - can you give me a quick pointer on how to get the true peak and LUFS values from it?

I suspect if you are hitting true peaks of -3dB with a DCP things might already be on the loud side ... but I think it should be safe from a clipping point of view.

[Wolfgang Woehl]

Media page (bottom, left-most icon) > add asset to project > Edit page > drag asset into timeline > Fairlight page > context click on asset > Clip operations > Analyze Audio Levels

[Wolfgang Woehl]

3. Would you recommend changing the gain in DCP-o-matic to get the True Peak to -3 dB, and can I rely on this measurement?

Hunting peaks at that point will make the whole mix too low - this should come clean from the sound people.

[carl]

I also checked the files in DaVinci Resolve. The True Peak readings are significantly different (approximately 1 dB of difference), although the Integrated Loudness (LUFS) values are quite similar. I've seen this behaviour with two different audio files.

Thanks, I see this too. The ffmpeg command-line tool gives the same result as Resolve on the random file I tried, so something seems definitely to be wrong with DoM.

I made a note in the bug tracker: https://dcpomatic.com/bugs/view.php?id=3176

[carl]

What sampling rate are the files you are testing? The discrepancy could be because of the resampling step that DCP-o-matic does to get things to 48kHz.

[Carsten]

When doing these tests, I would also compare sanple peak and true peak and see if the problem is with both values.

[macg]

I'm sorry for taking so long to reply and thank you for all the help.
After some testing I believe that the bigger discrepancies happen when there is resampling.
I also believe that the file I was using was not a good one and had some kind of distortion because both apps showed true peaks above 0db. With a better file I had just a 0.1db difference with resampling.
What is a bit weird is that after I change the channel layout the analysis get erased and the new one has different values.