GUIDE: How To Properly Encode Dolby Digital Audio (AC3)

[3ngel]

I wonder if Dts use this same (absurd in my opinion) cut frequency policy. Anyone tried it?

[3dsnar]

Hmm, not necessarily absurd. Since you can not hear such high frequencies, why to encode them?
(this is in gerenal the idea behing limiting the upper frequency band).
And this is somewhat true - most people cannot hear a sinosoid over 20 kHz, but AFAIK preserving higher frequencies is important for the harmonics of the sounds. I.e. cutting out harmonic components of the signal (even though they are over 20 kHz) affects the sound quality.

[3ngel]

Hmm, not necessarily absurd. Since you can not hear such high frequencies, why to encode them?

That's not the point. If i have a certain signal, i want that signal intact with all its frequencies, unless some freq limitation it's clearly declared (and as far as i know there is no freq limitation declared in the ac3 specifications).

but AFAIK preserving higher frequencies is important for the harmonics of the sounds. I.e. cutting out harmonic components of the signal (even though they are over 20 kHz) affects the sound quality.

That's the point

[raquete]

cutting out harmonic components of the signal (even though they are over 20 kHz) affects the sound quality.

it's right guys but low frequences have more audible harmonics....think in 60Hz and sum...now with 20K or more,just a few or "nothing"!
your amplifier/receiver maybe can answer more than 20K but the speakers...i can't trust.
we can listen 20Khz(when someone can) is "alone",not in music because have strong diference in low frequences(and middles) for very high frequences for our ears.20Khz is impossible to be listen in musics.

i want that signal intact

maybe you have that signal intact but you can't listen diferences with more than 20Khz(or a little less).
don't "bore" with too high(and inaldibles) frequences,pay double atention in the basses that have lots of harmonics and more volume in musics.basses are the "soul" of the quality,when it's good...sounds better!

ps:can someone take a look in the screenshots in my post on page 4 of this thread ( http://forum.doom9.org/showthread.ph...734#post840734 ) to remove my doubts about "normalizations" using sources with differents volumes?
( tebasuna51 (thanks again) send one on cool link that answer 99% of my doubts and only one more answer is needed)

thank you all

edit: typos

[tebasuna51]

Originally Posted by raquete

ps:can someone take a look in the screenshots in my post on page 4 of this thread ( http://forum.doom9.org/showthread.ph...734#post840734 ) to remove my doubts about "normalizations" using sources with differents volumes?
( tebasuna51 send one on cool link that answer 99% of my doubts and only one more answer is needed)

I don't answer before because I'm not sure about this. Take my comments only like a opinion. You have two chances with this kind of source (seems have a narrow Dynamic Range):

1) Make a Dolby compliant ac3.
The sources don't need to normalized before encode and use
source with 0dB DialNorm=-12 dB (Avg), DRC=Music Light
same source with -3dB DialNorm=-15, DRC=Music Light
Then the ac3 decoder can play your ac3 at same global volume than others Dolby ac3 contents.

2) Make a ac3 not Dolby compliant but to be played at same volume than others contents like mp3, CD Audio, TV, ...
Normalize to desired level and encode with:
DialNorm=-31 dB, DRC=None
Then the ac3 must be played at same volume than wav source.

[raquete]

edit: obsolent post,correct answer here: http://forum.doom9.org/showpost.php?...1&postcount=90

source with 0dB DialNorm=-12 dB ...same source with -3dB DialNorm=-15...Normalize to desired level and encode with

i'm sure you're right tebasuna51,but you show me the parameters of each audio to encode(as quoted)...i will do the tests.

what i really think is why SomeJoe use the preset
[Sys] Normalize RMS to -16 dB (music) using "scan levels"...?? see that he don't posted how much dBs had his source too,then, i still have doubts in the guide,not in the test that you propose.

if the result of your test isn't right or if is right(that i trust), the doubt remains the same...(why the -16 preset? because using any other preset give different result ...not -20)

thank you so much!

[tebasuna51]

@raquete
I'm not sure if I understand your question. Maybe...

1) When SomeJoe use Sound Forge "Normalization" feature is only to measure the RMS of wav source. The normalization is not applied (Cancel button) then the preset is indifferent.

2) Using Audition "Group waveform Normalize" you don't need go to step 3 to fix any normalization. Just in step 2 press the button "Scan for Statistical Information" and see the Avg value, this is your Dialog Normalization value for this wav source. Now you can "Close" the window.
The "*Percent over ... to -20 dB" (in red in your picture) is useless, you can put any value at step 3 "Normalize" and always the Avg value is the same ("Reset" and "Scan..." in step 2).

3) Your two pictures are coherent. First wav RMS Avg value -12 dB, second wave (3 dB below the first) RMS Avg value -14.76 dB. Not exact but coherent, different wav different RMS Avg value.

[raquete]

edit: obsolent post.correct answer here: http://forum.doom9.org/showpost.php?...1&postcount=90

@raquete
I'm not sure if I understand your question. Maybe...

sorry,i know...is my bad english,my fault.

1) When SomeJoe use Sound Forge "Normalization" feature is only to measure the RMS of wav source. The normalization is not applied (Cancel button) then the preset is indifferent.

....but he applied:

For this example, I will use an audio file that was captured from analog material. It is a plain stereo .wav file, 48 kHz, 16-bit. I loaded it into Sound Forge and measured the RMS level of the entire file as -20 dBFS.

Knowing that, here are some screen shots of the proper settings to encode this file...

The first page in ACID is the Audio Service Configuration, where the coding mode (2/0), the data rate (192 kbps), and dialnorm (-20 dBFS) are set.

here the screenshot od the "first page in ACID"

this is what i'm talking about.
using the "[Sys] Normalize RMS to -16 dB (music)" to find the normalization he got -20dB,using another preset(-10,-31dB or other) we get a completely different result
what i really don't understand is why -16dB preset was used to find the normalization (-20dB in this case) that was applied in the encoder (Acid) as show this screenshot from the guide.

do you know what i mean now tebasuna51?
thank you so much,(great guy).


edit: later i post the results using -16,-20,-27 and -31db in waves group normalize using the same source with same volume.
(all results are differents)

[raquete]

@
tebasuna51 and all

see what happens using differents values.

source: off the ground-P.McCartney
extracted from cd have -.23dB (97.39%) no norm or amplify.

group waveform normalize
adjusted in "normalize to a level of" "x" using "Equal Loudness Contour" (normalize tab)
and "scan for statistical information" (ananlyze loudness tab)
( Eq-Loud=-9.48 / Loud=-10.79 / Max=-4.94 / Avg=-12 / %Clip=0% )
of course always give the same result in the "analyse loudness" tab because is the same source (no matter what "x" value is choosed to scan)

loading the source 4 times,adjusting "x" values for each and results after run normalize :

-16dB= -6.75dB (45,97%)

-20dB= -10.75dB (29,01%)

-27dB= -17.75dB (12,96%)

-31dB= -21.75dB ( 8.18%)

as i "told you",for each value chosed give different result with the same source.
now think when you have differents sources/differents volumes to normalize.

so much.

[tebasuna51]

@raquete
Don't mistake:

1) Normalize a wav, with SounForge/Audition, is modify the amplitude (volume).

source: off the ground-P.McCartney
extracted from cd have -.23dB (97.39%) no norm or amplify.

group waveform normalize
adjusted in "normalize to a level of" "x" using "Equal Loudness Contour" (normalize tab)
and "scan for statistical information" (ananlyze loudness tab)
( Eq-Loud=-9.48 / Loud=-10.79 / Max=-4.94 / Avg=-12 / %Clip=0% )
of course always give the same result in the "analyse loudness" tab because is the same source (no matter what "x" value is choosed to scan)

Ok.

loading the source 4 times,adjusting "x" values for each and results after run normalize :

-16dB= -6.75dB (45,97%)

-20dB= -10.75dB (29,01%)

-27dB= -17.75dB (12,96%)

-31dB= -21.75dB ( 8.18%)

as i "told you",for each value chosed give different result with the same source.

"after run normalize", for what?
You have different wav with different volume then you have different RMS Avg value.

2) Calculate the DialNorm to ac3 Dolby compliant encode. You need only know the RMS Avg of your wav source. The amplitude is not modified, only the parameter DialNorm is added to the ac3 stream.
Then "...but he applied:" don't mean the volume is modified (like Normalize) only the parameter DialNorm is set to -20 dB

[raquete]

"after run normalize", for what?....
You need only know the RMS Avg of your wav source. The amplitude is not modified, only the parameter DialNorm is added to the ac3 stream.

oh boy,only now i understand,...you're completely right and the guide too of course.
now i'm very embarassed: what i do with my obsolents posts here?!? (maybe one advice is needed in each showing that they are obsolents with a link to your last post)

to SomeJoe and triple for you tebasuna51 that really (with big patience) help me to understand it all and remove all my doubts.

beers for you and SomeJoe, you are very cool!

[smok3]

quote from http://etvcookbook.org/audio/dialnorm.html :

"The value of the dialnorm parameter in the AC-3 elementary bit stream shall indicate the level of average spoken dialogue within the encoded audio program."

spoken dialogue, mkay? (as someboy noted before that rms calculus for the entire file is not valid...)

also, afaik rms is not a really good value for subjective loudness, there are better algos out there, like replaygain.

http://www.replaygain.org/

edit: guessing, lets say you have a 2 channel mix that you want to encode into 2ch ac3:
cut out the dialogoues, merge them and then calculate the rms (or whatever value, again RG should be better if we could match the values.) and set the dialnorm based on that.

edit2: another thing, dynamic compression: iam pretty sure this is used to prevent clipping from lossy source (like ac3) as well, and not only for subjective purposes (dolby does suggest to actually mix the audio throught the encoder, that may be one of the reasons why, right?)

[raquete]

@ tebasuna51 or anyone that can help me more if possible:

For this example, I will use an audio file that was captured from analog material. It is a plain stereo .wav file, 48 kHz, 16-bit. I loaded it into Sound Forge and measured the RMS level of the entire file as -20 dBFS.

all right.SomeJoe found -20dBFS but don't tell us how much volume had his source.
what i want to mean is that each source (differents volumes) will give differents RMS levels.
my first doubt:
how much dBs have to have the source? (before measure the RMS level)
second and deep doubt:
SomeJoe used one stereo source,then,what can i do to encode 5.1 sources (meaning L,R,C,LFE,LS and RS tracks),how much volume have to have each track?

i'm sorry for my complicated (maybe unclear) questions

thanks.

[ashp8]

i use wavelab5, nero wave editor(nero 7), Nero sound trax, maved3d and goldwave to edit my source material to encode i use sonic foundry soft encode 1.0.

I understand all licensed dolby digital encoders are the same in output, they have configuration and each configuration can be applied like in one encoder axactly same in the other another.

I've messed around with centre channels and resolved my issue of quiet or unnoticable centre channels by altering the audio in wave editor by volume and applying a compressor in goldwave to limit loud audio. this allows me to get a loud clear vocals channel in the centre.

Now my issue is using different programs they work at different bit depths and my audio sounds choppy as i boost the volume using the controls inside sonic foundry soft encode. if i lower the volume then it becomes less bright and dull and not loud enough.

i hate dolby digital. it is not for Music it lowers the centre and surround channels despite the dialog norm. volume management and it is very picky as the audio must be edited once and in 32-bit finely dithered to 16bit. DTS is much better i can pre-master the audio to make sure it is not too loud or quiet i keep my audio at -1.5db max and all centre, rear and front channels are balanced why isn't dolby digital music friendly.

how do i make sure that if i decrease volume there are absolutely no choppiness in audio i am confused.

[tebasuna51]

Originally Posted by raquete

my first doubt:
how much dBs have to have the source? (before measure the RMS level)
second and deep doubt:
SomeJoe used one stereo source,then,what can i do to encode 5.1 sources (meaning L,R,C,LFE,LS and RS tracks),how much volume have to have each track?

1) The source volume is not a requirement to encode to ac3. Use the volume you want (maybe normalized to 95%).

2) Keep the relative volume between original channels (the word 'track' is used to the whole ac3: english audio track, spanish audio track, ...). Don't normalize each channel, use the same gain for all channels.

[raquete]

the word 'track' is used to the whole ac3..

ok,clear....but what about dialnorm?(later we talk about this)

Don't normalize each channel, use the same gain for all channels.

but what about if i'm using one source with ~95% and extracting all channels?
what i want to mean is that each channel will have:
LR ~95%(source)
C(center only) ~95% (depend of the source)
LFE ~25%(i use discrete amplifier only for lfe)
SLSR(surrounds less center) ~95%(depend of the source too)
total volume mixing the channels = "hundreds" %

the big question is: how much have to have each channel

thanks tebasuna51.

[SomeJoe]

Originally Posted by craftech

It is hard to believe this post hasn't drawn much criticism yet. There is absolutely no way that the method of determining dialnorm at the start of this thread will produce anything but a muffled, muted, and terrible sounding audio movie track.

The reason is simple. The standard for dialnorm was not based upon the overall soundtrack, it is based upon the dialogue. The simplistic method of calculating it described at the start of the post cannot work because the original posted doesn't understand the basis for dialnorm in the first place.

My apologies for not answering this post sooner, I have been busy in the past months and don't login and read here as often as I used to.

Craftech is somewhat correct in his statement, that dialnorm is supposed to be based on the level of the dialogue, not the entire soundtrack. I did not make this clear enough in my original guide, and for that I apologize.

However, I take exception to the statement that I don't understand the basis for dialnorm and the statement that says that this method can't produce good audio. Allow me to explain.

First, my apologies for coloring the original guide with my own experience. My company produces primarily educational DVDs that are predominantly seminars and classroom-type courses. As such, the entire soundtrack of these DVDs is dialogue, with virtually no music or sound effects. Measuring the RMS level of the entire soundtrack in this case results in a value that closely correlates with a proper dialnorm setting.

In many other instances, the soundtrack will contain a mixture of dialogue, music, sound effects, and other content. In these cases, you should indeed measure only a portion of representative dialogue when attempting to determine the RMS level of the audio to be used for setting dialnorm. This is accomplished easily enough in Sound Forge (or other software) by selecting only the section of audio with dialogue before measuring the audio with the RMS/Normalization tool.

As to the other concern recently posted in this thread, that RMS is not a good measure to use because it doesn't correlate well with LAeq, I already addressed this in the original guide. I have previously stated that RMS measurement tools, because they are available in easy-to-obtain software, provide a "poor man's" method of obtaining a value close to the real LAeq level that dialnorm is supposed to be set with. I am well aware that RMS is not perfectly correlated with LAeq, but as I (nor anyone I know) owns software or hardware to measure LAeq directly, RMS will have to serve as a substitute.

There have been some suggestions in this thread that other methods other than RMS may correlate more closely with LAeq. I have not investigated that, and have no ability to do so. Of course, should someone have some evidence that another method would work better than RMS, then by all means use it, and post your results here. (In fact, if you have software or hardware to measure LAeq of your dialogue, do that).

As further evidence that reinforces my belief that RMS measurements will suffice, I offer this: I applied a few years ago with Dolby for use of the Dolby Digital logo on my company's DVDs. The Trademark and Standardization agreement required that I submit samples of my DVDs to Dolby for approval of the method. The first samples I sent were rejected because dialnorm, DRC, and a few other parameters were not set correctly in my encoded AC3. After research in the previously mentioned/linked Dolby documents, I came up with the method I posted in the guide, and I resubmitted my DVDs to Dolby for approval. They came back approved this time. Since Dolby's approval processes are considered to be somewhat rigorous, I can only conclude that my method, even though RMS sometimes does not correlate well with LAeq, arrives close enough to correct parameters to pass Dolby's approval processes.

In the interest of being correct, I will modify the guide to emphasize that the RMS level should be computed from a dialogue portion of the soundtrack, not the entire soundtrack.

[tebasuna51]

Originally Posted by raquete

...
total volume mixing the channels = "hundreds" %

the big question is: how much have to have each channel

What mix? Each channel is encoded separately (more or less), then all channels can be at 100% of volume (not habitual but possible).
If an ac3 5.1 is played by a stereo equipment must do a downmix with a normalized matrix to avoid saturation problems.

[raquete]

What mix?

i mean "mix" as encoding AC3

If an ac3 5.1 is played by a stereo equipment must do a downmix with a normalized matrix to avoid saturation problems.

....of course,understood.

Each channel is encoded separately (more or less), then all channels can be at 100% of volume (not habitual but possible).

all right.
what do you think if i use each extracted channel(C,SLSL and LFE(i mean sub-woofer)) without amplification? (for example,if the extracted center channel have 70% of volume from source(LR))
i'm asking because using all channel ~95% give me too loud volume using 6 discrete amplifiers(not receiver/HT)

thank you tebasuna51.

[tebasuna51]

Originally Posted by raquete

what do you think if i use each extracted channel(C,SLSL and LFE(i mean sub-woofer)) without amplification?

If you extract the channels from a original ac3 is recommended don't modify the volume levels. But you are free to make any personnal touch.