User17809199528284647324 (Customer) asked a question.
Unexcplicable Loudness Measurements and Behaviour
Hi,
Because of the extremely high, ear-deafening loudness in many movies, I started investigating the loudness characteristics of these soundtracks. I initially assumed that the departments responsible for consumer releases were the cause, but my measurements seem to indicate that there might instead be an issue in the encoding/decoding process.
Can these measurements be explained, or is it possible that there is a bug in the encoder?
After extracting the AC-3 and DTS audio tracks from my Avatar 1 3D Blu-ray disc, I measured their loudness using Youlean Loudness Meter.
Since Youlean cannot decode DTS directly, I first decoded the DTS stream into individual WAV files and then combined them into a single 5.1 WAV file in Adobe Audition.
The loudness histogram of the AC-3 track looked completely different and appeared heavily distorted, with an extreme dominance of very loud levels, whereas the DTS-derived 5.1 WAV looked perfectly normal. Therefore, I encoded the DTS-derived 5.1 WAV using Dolby Media Encoder (v3.7.0).
I selected the Dialog Intelligence option and measured with a 15% cutoff.
The first image shows the resulting loudness measurement in Youlean. The loudness distribution looks almost identical to that of the original AC-3 5.1 track. However, on the left side of the image, you can also see the loudness histogram of the DTS-derived 5.1 WAV.
This histogram looks normal and healthy, with lower levels dominating throughout most of the movie.
To rule out an issue with Dolby Media Encoder, I also encoded the DTS-derived 5.1 WAV using the older Minnetonka SurCode Dolby Encoder. The result was essentially identical.
Next, I used the freeware tool eac3to to convert the same 5.1 WAV source to AC-3 without specifying a dialnorm value (MediaInfo reports the resulting file as having a Dialog Normalization value of -31 dB).
In the second image, you can see the loudness distribution of the AC-3 file created by eac3to and decoded directly in Youlean.
This time, the histogram looks correct and closely matches the original DTS-derived audio (left side of image #1).
Any insights into what might be causing this behavior would be greatly appreciated.
Julius
Because of the extremely high, ear-deafening loudness in many movies, I started investigating the loudness characteristics of these soundtracks. I initially assumed that the departments responsible for consumer releases were the cause, but my measurements seem to indicate that there might instead be an issue in the encoding/decoding process.
Can these measurements be explained, or is it possible that there is a bug in the encoder?
After extracting the AC-3 and DTS audio tracks from my Avatar 1 3D Blu-ray disc, I measured their loudness using Youlean Loudness Meter.
Since Youlean cannot decode DTS directly, I first decoded the DTS stream into individual WAV files and then combined them into a single 5.1 WAV file in Adobe Audition.
The loudness histogram of the AC-3 track looked completely different and appeared heavily distorted, with an extreme dominance of very loud levels, whereas the DTS-derived 5.1 WAV looked perfectly normal. Therefore, I encoded the DTS-derived 5.1 WAV using Dolby Media Encoder (v3.7.0).
I selected the Dialog Intelligence option and measured with a 15% cutoff.
The first image shows the resulting loudness measurement in Youlean. The loudness distribution looks almost identical to that of the original AC-3 5.1 track. However, on the left side of the image, you can also see the loudness histogram of the DTS-derived 5.1 WAV.
This histogram looks normal and healthy, with lower levels dominating throughout most of the movie.
To rule out an issue with Dolby Media Encoder, I also encoded the DTS-derived 5.1 WAV using the older Minnetonka SurCode Dolby Encoder. The result was essentially identical.
Next, I used the freeware tool eac3to to convert the same 5.1 WAV source to AC-3 without specifying a dialnorm value (MediaInfo reports the resulting file as having a Dialog Normalization value of -31 dB).
In the second image, you can see the loudness distribution of the AC-3 file created by eac3to and decoded directly in Youlean.
This time, the histogram looks correct and closely matches the original DTS-derived audio (left side of image #1).
Any insights into what might be causing this behavior would be greatly appreciated.
Julius
You made a lot of work with this analyze.
However I would start that you should analyzy it with the original files as an input and check what is happening then.
Here you have started from ripping BluRay - which is not allowed due to copyrights.
You haven't specify what tools you have used as decoders for taking back AC3 in to pcm wav files and what were the settings for that. Did your decoder took properly metadata from Dolby Digital bit stream. Please note that Dolby Digital is a lossy codec so the result od decoding will be different then the original files used for encoding by the studio.
Make sure also that your settings for decoder while you are listening to the origianl BluRay disc are set properly.
Could you also please let us know what is your target use case here and what is the main reason of this analyze?
>However, I would suggest analyzing it using the original files as input and checking what happens in that case.
That is exactly what I did, using a 5.1 WAV mix as the reference.
The first screenshot shows the loudness histogram of the original on the left and the encoded material on the right.
> You haven't specified which tools you used to decode the AC-3 streams back into PCM WAV files, nor what settings were used.
I did: Youlean, Adobe Audition, and XRECODE 3.
It all started because I assumed there might simply be something wrong with my setup. I double-checked everything and verified all measurements. Everything appeared to be fine.
Then I decoded an AC-3 stream in Audition, and I could hardly believe what I was seeing: the center channel waveform was completely pushed against the ceiling, heavily limited like a modern pop song.
I was genuinely shocked by the apparent incompetence of those responsible for the consumer version, who seemingly do not even understand that limiting does not reduce loudness—it only reduces peak levels.
I checked several movies containing AC-3 streams, and several of them showed a heavily limited center channel.
I then researched further and found Youlean for more detailed loudness analysis, as it can decode AC-3 directly and offline.
It also showed extreme clipping and roughly the same levels as Audition.
After that, I searched for additional alternatives and found XRECODE 3, which, to my surprise, decoded the AC-3 stream into WAV files at lower levels and with significantly less clipping. However, the decoded 5.1 WAV still exhibited the same oddly skewed loudness histogram that I observed when decoding the AC-3 stream directly in Youlean.
I then performed several tests using Dolby encoders.
I used extreme Dialnorm settings of -1 and -31.
At that point, I noticed another very strange behavior:
Although Dialnorm was set to the lowest playback level of the decoder, Youlean showed that the true peaks were reduced far less than would be expected from a simple reduction of the decoder output level.
Even though the peaks were no longer clipping at the lower Dialnorm setting, the shape of the altered loudness histogram remained distorted and unchanged. The entire distorted histogram was simply shifted downward to lower levels, but it never resembled the loudness histogram of the original uncompressed WAV file (verified with Audition's loudness distribution measurement).
How is it possible that, when encoding with a Dolby encoder, both Youlean and Audition show a complete distortion of the loudness structure, while encoding with eac3to results in Youlean and Audition showing an unchanged loudness structure that remains identical to the source material?
In my opinion, this is strong evidence that either a parameter is being set incorrectly during encoding — but int that case obviously not only by me, but apparently also by consumer release departments — or that there is a bug in the encoder itself.
Further questions arise from these measurements:
When a 5.1 AC-3 or Dolby Digital Plus stream is generated with Dolby Media Encoder and the dialogue level is measured by DME in order to determine Dialnorm correctly, how is it possible that the resulting AC-3 stream and DD+ stream, when decoded in Youlean, exhibit true peaks way above 0 dB, different loudness values, and a different loudness histogram compared to the original 5.1 source material?
That should not happen.
Also the encoded stream is supposed to sound identical to the original, and only when Dynamic Range Control (DRC) is enabled during playback should the dynamics be altered at all.
However, all of my tests indicate that the resulting Dolby stream acquires a different dynamic structure during the encoding process.
This behavior occurs with both older and current Dolby encoders.
>Could you also please let us know what is your target use case here and what is the main reason of this analyze?
I do not see how my motivation could invalidate the measurement results, but I guess my reasons for investigating this issue are threefold:
- Protecting hearing from sound levels that are simply unacceptable. When content is marketed as having "cinema sound," yet produces levels in much smaller listening environments that would be excessive even in large theaters, this issue should be addressed—and addressed urgently.
- A technical interest in determining whether I am making a mistake somewhere or whether there may be an encoding or decoding problem. After so many years of discussion and the establishment of loudness standards, countless films still contain minute long screams, crashes, and shattering sounds exceeding 95 dB(A!!!). If measurements performed with Youlean confirm extreme loudness in the source material, perhaps the root cause of this persistent (loudness) problem in (consumer release) movies lies precisely where hardly anyone has considered looking: in the encoding software that has become the industry standard?
At a time when many people in the industry are concerned about job security, even fewer are willing to question established assumptions critically.
- Finally, as a producer, when I subjectively hear that something cannot possibly be right, and subsequently find measurement data that supports that impression, I believe it is worthwhile to investigate further. If there is indeed an underlying cause, perhaps identifying it could help eliminate the problem once and for all.
I would like to confirm first one thing. In your first post you have mentioned you are using BluRay extracted Dolby Digital bitstream. You have decoded it using 3rd party decoder to wav. Right?
By the original files, I mean those from studio, and I believe you do not have access to the original Dolby Atmos Master file used for BluRay encoding job.
You have also mentioned that you have similar findings to different BluRay AC3 streams.
Because of using open source software, that is mostly not using Dolby Certified components, your investigation might be not 100% accurate and inconsistent to compare different formats - especially that for each one you are using different tools.
Could you also please share your log files from Dolby Media Encoder in private message, or please open w case and attach them? You can find email in the help section of Dolby Media Encoder.
Regarding Youlean loudness meter, manual is not explicitly mentioning support for Dolby Digital format, it is mentioned more or less in the group of a system codecs. Also does it allow to set decoder parameters, does it respect metadata as Dialnorm?
Many thanks.
Why are you now fixating on a Dolby Atmos master file?
I extracted the AC-3 streams and had them decoded by both Youlean and Audition.
Encoding WAV files with both DME v3.7 and SurCode produces an AC-3 stream which, when decoded by Youlean and Audition, exhibits the same loudness histogram and true-peak issues.
>Regarding Youlean loudness meter, manual is not explicitly mentioning support for Dolby Digital format, it is mentioned more or less in the group of a system codecs. Also does it allow to set decoder parameters, does it respect metadata as Dialnorm?
Both programs do not allow to select DRC or make any changes before decoding starts.
If even these professional programs show these problems, how many home theater receivers or consumer devices will end up with the same decoding result?
Yes, I did more tests with different 5.1 mixes but the result strangely was always the same: the loudest cumulative levels in the histograms of the original mix, which - naturally - are not the most often occurring levels, after encoding and decoding are shown as occurring most often.
In the newly attached picture of the metadata of the AC3 stream (shown by MediaInfo), it can be seen, that Dialnorm by DME was set low enough (-11), to prevent any limiting effects on any channel - and still the loudness histogram of the decoded AC3 file looks completely different from the original's loudness histogram.
I also include the DME log file to check for the used encoder settings - I do not see any incorrect setting used, that would explain, why the loudness histogram becomes changed at all and looks so different from the 5.1 source.