dBFS audio measure

[Goku193]

Hi,

I'm preparing a master DCP package of a short film for a screening at a upcoming international film festival. I've received a list of specifications from the festival and two questions popped into my head while reading it.

1) The festival recommends that the audio tracks have Reference level: -20dBFS. I've read that -20 dBFS is a SMPTE standard and is equivalent to 85 dBSPL, but I was wondering how to check if the audio level is right through DCP-o-matic (the LEQ(m) ?).

I'm attaching two audio graphs. The Source.png is the audio graph of the copy that I've received from the filmmaker, and the Normalized.png is the audio graph of the copy with normalized audio to -20 LUFS and Peak of -1.1 to avoid clipping.

I've also ran a FFmpeg volumedetect filter to both files - this is the results:

Source file:
[Parsed_volumedetect_0 @ 0000022275cd8a80] n_samples: 78328320
[Parsed_volumedetect_0 @ 0000022275cd8a80] mean_volume: -23.3 dB
[Parsed_volumedetect_0 @ 0000022275cd8a80] max_volume: 0.0 dB
[Parsed_volumedetect_0 @ 0000022275cd8a80] histogram_0db: 4234
[Parsed_volumedetect_0 @ 0000022275cd8a80] histogram_1db: 17607
[Parsed_volumedetect_0 @ 0000022275cd8a80] histogram_2db: 32575
[Parsed_volumedetect_0 @ 0000022275cd8a80] histogram_3db: 42442

Normalized (-20 LUFS).
[Parsed_volumedetect_0 @ 000002117a048a80] n_samples: 78328320
[Parsed_volumedetect_0 @ 000002117a048a80] mean_volume: -22.2 dB
[Parsed_volumedetect_0 @ 000002117a048a80] max_volume: -1.1 dB
[Parsed_volumedetect_0 @ 000002117a048a80] histogram_1db: 1882
[Parsed_volumedetect_0 @ 000002117a048a80] histogram_2db: 6719
[Parsed_volumedetect_0 @ 000002117a048a80] histogram_3db: 15715
[Parsed_volumedetect_0 @ 000002117a048a80] histogram_4db: 31200
[Parsed_volumedetect_0 @ 000002117a048a80] histogram_5db: 57368

I would appreciate any hints about what to look at and if the graph doesn't show any potencial problems with the audio. I will also check the audio level at the cinema, but I wanted to ask about how to check if this reference point is meet.

2) This is a simpler matter: according to guideline in DCP 2.0 Channel tracks (stereo) have to be mapped on 5.1. As seen on the graphs, the film is in Stereo, but I'm assuming they asking about making a 5.1 DCP only with L and R channel used. Does anyone meet with such requirements and can confirm it?

Thank you!

[Carsten]

Around -20 LUFS looks okay to me. When festivals mention reference levels, they typically only repeat what is common in cinema mastering. It is not their specific festival parameters. Just as festivals can not control creative aspects of the imagery of submissions, they are not to control the loudness of an audio mix. It's a creative decision. What they want to make sure though, is that the mix is within a no-complaint tolerance area.

When I look at these two graphs, it is apparent that not only normalisation has been applied, but also some dynamic expansion. Both mixes are around -20 LUFS, but the source has a much lower dynamic range (13.46LU vs. 23.05LU). This is also reflected in the RMS curves. How did you achieve this?


Regarding stereo vs. 5.1 - yes, common practice is to use a 5.1 track, with only the first two channels being poulated with an L/R mix. You could try to use the LCR Upmixer (or any other external Upmixer), but the filmmaker would have to pass this mix before submission.

Default behaviour/settings in DCP-o-matic is a 5.1 DCP, and if you add a stereo track, it is routed to L/R only.

I have seen many smaller festivals supplying only stereo playback systems. It might be worth inquiring wether they use proper cinema auditoriums with a dedicated center speaker, otherwise, the center mix may be wasted or could even cause problems. Sometimes, festivals may demand stereo mixes explicitly for that reason.

[Goku193]

Dear Carsten,

thank you for the reply. I understand that reference levels is only the estimate to make sure that the audio won't cause any issues, but I also wanted to make sure that it would sound right.

Regarding the dynamic range: I've used the loudnorm EBU R128 filter of the ffmpeg with the set values for loudness (-20) and True Peak (-1.1). The default value of dynamic range (LRA) is 7.0, so I'm assuming this is why it's so different. Do you think it would be better to just drop the left channel by -1 dB of the master with higher loudness range (wouldn't be too quiet for the cinema though?) or use the normalized audio?

For the Stereo audio and 5.1 track, I will just stick with L/R mix, but thank you for the confirmation.

[Carsten]

Again, this just an educated guess based on these graphs - the source version with everything hovering around -20dB RMS would probably be perceived as too loud. The normalised version looks better with average RMS between -30 and -20dB - however, I still don't understand how you achieved that. I hope these graphs are reliable, as a straight simple normalisation can not accomplish that. But - you should be fine.

[Goku193]

Dear Carsten,

thank you for the reply. I think that you were looking at the graphs in reverse order. The source version has bigger loudness range, and the normalized version is 13.46 LU, while maintaining the -20 LUFS value.
I was doing some tests with different values for the normalisation (setting the targeted LRA), trying to maintain the Loudness range and not change it so drastically. I also made a simple -1dB change in gain. I'm attaching the graphs in the archive, not to spam so many images - they are named based on the used settings.
I'm thinking of using the 19LUFS_9LRA audio. Do you think it could be perceived too loud and it's better to just stick with the source audio dropped by -1dB?

[Carsten]

Which tool did you use for normalising?

At -19 to -20 LUFS, both versions should be fine as far as complaint-level go. Wether a version with reduced dynamic range is better or not, I can not judge without listening to it.

The normalised version has a higher overall RMS, around -20dB over the first two thirds of playtime. That could be too much for a 12min short - but it still depends on the type of material. What type of audio is it?

[Goku193]

I used the loudnorm filter in ffmpeg: https://ffmpeg.org/ffmpeg-filters.html#loudnorm

The first half of the film is filled with ambient music on the background and voiceover of the hero who is moving through a crowded city and goes into the club (in the half of the film - that's the biggest spike) and after that we have some dialogues and more ambient music. It's all made in DIY style.
I can send you the link to audio, but I can't send private messages yet, so if you want to check it and send me your contact, that would be great.

[Carsten]

I see. That loudnorm filter does not just apply plain linear/peak normalisation, but also allows to apply dynamic compression. That explains why dynamic range also has been altered. From my judging of the curves, I would assume that the original/source version is better, the normalised version may expose too little variation in dynamics and the average RMS level may be too high. But that's just guessing.

- Carsten