Need advice about mixing 5.1 down to stereo

[Cornucopia]

@hello_hello, those are nice try posts and all, but...

those are just inclusionary lists of tracks, not the actual ordered track assignments.

Here you go:
DTS/mpeg2 - ID#1 from Table 3-2, p20
DTS/iso/mp4 - ID#31 from Table 2-4, complex bit masks...
DCI - right there in middle of the single page
AES - Table 5 from p18
EBU - Table 3 from p8
ITU - Table 2 from p3
FLAC - FrameHeader table, #4ch assignment = 6 channels "follows SMPTE/ITU"
WAV - "Default Channel Ordering" section
Dolby - Table 4-1, p4-1
SMPTE reference didn't transfer correctly (members-only link), but here is Aussie Broadcasting Corp, referring & conforming to it:
#7.10 Track assignments for DolbyE (AC3 follows this as well), on p7.
http://www.abc.net.au/tv/independent/doc/ABC_Delivery_Specs_Aug_2011.pdf

I didn't duplicate, officer, I swear! I just passed along my copy. It's like a loan.

I also don't see the likelihood of unique LFE-only channel effects happening much. As was mentioned in a couple of those recs as well as a few more from Dolby and "Grammy Producers", it is often unused now due to the fact that digital can handle full range tracks in all the other channels, but was left in for 1:1 legacy material transfer.

Scott

[hello_hello]

Djard,
I don't suppose you use foobar2000 do you? Because I discovered this today, which actually annoyed me a little as I had no idea it existed. Foobar2000 contains a downmix DSP but it's set. You can't see or change what it's doing. This third party DSP is fully configurable. In the past I've being doing non-standard downmixing (re-mapping channels, changing their volumes etc) by decoding via directshow and using ffdshow's mixer filter. It works fine, but I could have achieved the same thing without as much messing around. Anyway.....

The foobar2000 Matrix Mixer DSP can be found here: http://skipyrich.com/wiki/Foobar2000:Matrix_Mixer
If you're not familiar with foobar2000, you can add DSPs to the playback chain, configure the configurable ones, and save your DSP setup as a preset. Foobar2000's converter can also load the same DSPs and the same DSP presets. It's converter can also save conversion setups as converter presets.
The upshot of all that is.... currently you're importing 5.1ch audio into Audacity, doing whatever you're doing to the channels levels, exporting that as a stereo file, then maybe, converting that to anther format.

Alternatively you could load an AC3 audio file into foobar2000, right click, select an appropriate conversion preset, and the 5.1ch audio will be downmixed to stereo using your pre-configured mixer matrix as it's being converted. Much, much faster. Here's what the foobar2000 Matrix Mixer looks like when you configure it. I just applied a quick "standard downmix" configuration for testing. It works as advertised.
The "normalise" checkbox seems to reduce the downmixed audio by a preset amount. I guess to prevent the clipping I've been harping on about. I haven't worked out exactly how much it reduces yet. Or you can set your own multiplier which I assume also adjusts the volume of the downmixed audio by a preset amount.



Edit: And after a little more playing around and Googling, I came across this. Another DSP I might have used on occasion had I know it existed: Winamp DSP Bridge

So it turns out the Winamp DSP plugin allows foobar2000 to use Winamp DSPs, and the RockSteady plugin I use for compression on playback happens to be a WinAmp DSP. It didn't take me long to configure another foobar2000 conversion preset to include the Matrix Mixer, followed by the Winamp DSP Bridge running RockSteady, and to save that as a new conversion preset.
Now I can load 5.1ch audio into foobar2000, right click, select convert, choose the appropriate preset, and the 5.1ch audio is downmixed to stereo using the Matrix I specified, compressed by RockSteady, and re-encoded in a single step.
It's pretty fast, even using this old PC. A one hour, 5.1ch AC3 file can be downmixed to stereo, compressed and converted to FLAC in about a minute and a half.

[hello_hello]

@hello_hello, those are nice try posts and all, but...

those are just inclusionary lists of tracks, not the actual ordered track assignments.

Why on earth wouldn't they display tracks in a real order?

Here you go:
DTS/mpeg2 - ID#1 from Table 3-2, p20

A table showing "Original audio configuration" and "Original channel count"?

Dolby - Table 4-1, p4-1

The way I read that it's referring to the recommended "input" channel order. It follows the 5.1ch WAVE order.

Table 4.5 on page 4-14 shows the "audio encoding mode" It specifically says:
"The Audio Coding Mode (also referred to as Channel Mode) parameter defines the number of main audio channels within the encoded bitstream and also indicates the channel format."

It then shows the channel format as this: L, C, R, LS, RS

SMPTE reference didn't transfer correctly (members-only link), but here is Aussie Broadcasting Corp, referring & conforming to it:
#7.10 Track assignments for DolbyE (AC3 follows this as well), on p7.
http://www.abc.net.au/tv/independent/doc/ABC_Delivery_Specs_Aug_2011.pdf

I didn't duplicate, officer, I swear! I just passed along my copy. It's like a loan.

Well I can see it appears the ABC requires Dolby-E to use a "standard" channel layout, although even that's a little ambiguous:

7.10 The track assignment of the Dolby E stream shall conform to SMPTE 320M as follows:
Track 1 Left
Track 2 Right
Track 3 Centre
Track 4 LFE
Track 5 Left Surround
Track 6 Right Surround
Track 7 Lt or freely assigned
Track 8 Rt or freely assigned

7.11 Track identification (during line up test signals) within the Dolby E stream is preferred, but not
mandatory if conformed to SMPTE 320M

Track assignment shall conform to SMPTE 320M, but if it doesn't, use track identification?
Anyway, the "Dolby DP-569 Multichannel Dolby Digital Encoder" was mentioned, so I found a copy of the manual for it. There's inclusionary channel assignments everywhere. Even on the front panel it seems.



And I'm not quite getting my head around why you'd set inclusionary channel assignments in the input/output configuration, but maybe I'm not understand the channel assignment function.



It appears the channel assignments can be changed, and it'd appear the ABC requires SMPTE 320M assignments, unless track assignments are specified, when they still do, or they don't, or however that works, but I'm still not seeing anything which shows the channel mapping in the AC3 audio I extracted from the vob files on my DVD is going to be the same.

[hello_hello]

Is the default input track assignment listed here the default order, or is it just inclusionary?

[pandy]

Nope. There's ALWAYS power in the audio world - it has to get to your ears! That is going to affect how you mix. You don't have new insight about this duality, it has been around for decades.

? or you talking about acoustic or electric signal representation - power imply particular current and and particular amplitude (level) - do you know your current (it assume to use specified impedance).
Unless it is clearly specified then we talk about level not power. Dolby use only level reference in their acceptance testing, sometimes they use analog signal units but then impedance is specified clearly (so current and power can be calculated but they are not relevant from acceptance perspective).
Inside audio stream (can be PCM can be AC-3 can be something else) there is only level (expressed as 0..1 or as +-1 or as absolute integer value) but there is no power - power will be created at the loudspeaker input.

And personal audio (when it's done right) is based on professional/broadcast audio.

Scott

To simplify our discussion it is same principle as setting treble, bass or more refine frequency equalization perhaps with DSP in home hifi equipment. Some people will put bass and treble to maximum some not, some prefer plain vanilla characteristic some not - subjective and personal, for objective use provided by Dolby downmixing equation (used from at least 10 - 15 years without change).

[hello_hello]

I found this. HTML5 AAC Audio Playback Tests - Multichannel

Which I still don't fully understand given it's AAC but uses the same channel mapping as AC3 (at least for the 5.1ch audio) according to what's written on the page. Maybe the HTML5 specification uses the AC3 channel layout? I have no idea. I did try a few samples. I only have stereo speakers but the appropriate output meters were bobbing up and down in foobar2000 so it was getting the channels right.

Aren't Fraunhofer a little involved in the development of AAC? Because they have this to say about AAC channel mapping.

The AAC codec family has supported up to 48 channels of audio since its initial development through predefined channel configurations and a flexible escape mechanism. The predefined channel configurations from the 2005 version of the AAC standard are:
(for 3 channel audio) Center, Left, Right
(for 4 channel audio) Center, Left, Right, Surround
(for 5 channel audio) ITU BS.775-1
(for 6 channel audio) ITU BS.775-1 (5.1ch)

https://www.itu.int/dms_pubrec/itu-r/rec/bs/R-REC-BS.775-3-201208-I!!PDF-E.pdf

Page 6:
Three-channel stereo plus 2 surround
L/C/R/LS/RS

FFS! That's the same as AC3 again. Does AAC really put the centre channel in a different order depending on the number of channels? If the above is correct, the consensus as to the channel order for 5.1ch AAC on this page must be incorrect.
I think I'll give up. Unless..... does anybody know how to look at multichannel audio..... AAC, AC3, FLAC, DTS etc and determine the actual channel mapping used? Can it be done via Avisynth? There must be a way. I tried using Avisynth and GetChannel() but I'm not sure how to determine whether the decoder is re-mapping the channels.
According to the info on the Fraunhofer page, the centre channel is channel 2 in their samples, but the only way I can get just the centre channel is with GetChannel(3). Which seems to apply no matter what type of audio I try, so maybe ffms2 is re-mapping when it decodes?

[hello_hello]

I did some conversing with QAAC via the commandline. Here's QAAC's opinion on the AAC channel layout.



Then I took a standard 5.1ch AC3 file and re-encoded it as a multi channel wave file using foobar2000, and when it was done I asked QAAC to convert the wave file to AAC for me.



QAAC shows the input channel layout as being the standard 5.1ch WAV layout as expected. It also shows the output layout as being the same as the AAC 5.1ch layout I linked to originally. http://avisynth.nl/index.php/GetChannel

It seems the AAC output channel layout is not the standard WAV 5.1ch layout, it's the mpeg channel layout mentioned previously. The same layout used by DTS, apparently.

This is a few years old, but.....
[qaac] current status of multichannel encoding
shortly after:
[qaac] release 0.77
"qaac has now implemented much of channel mapping; It does not only rearrange to AAC order, but also map channels to be coherent with QT layout (like SL, SR -> BL, BR).
Therefore, even if you are using older CoreAudioToolbox.dll, you are now encouraged not to use --native-chanmapper."

There seems to be a lot of channel remapping going on if all the various formats use the same mapping.

As per my previous post, I'm still a little confused regarding WAV vs AC3 channel mapping, when they appear to be different, but I've found references to them being the same. Such as nu774's reply in this thread:
https://github.com/nu774/qaac/issues/18
"No, qaac assumes inputs are in usual WAV/AC3 order."

I still don't understand why the AAC files I linked to in my previous post appear to have an AC3 channel order, although looking again I think maybe that's just the order the channels are heard, and not the actual encoder channel mapping.

[Djard]

I find that thinking logically is a handicap when consulting manuals in the comain of computing. But ooh! You've got me scrambling for the manual to my AIWA 5.1 system. Hopefully I'll find a descriptor for 'night mode'. Thanks.

I dragged and dropped the movie Popeye (1980) into Audacity and juxtaposed the six tracks. No question in my mind that the LFE track has some unique frequencies. I am even able to create them by taking a sample from CH-1 and using the EQ to drop the pitch by an octave, and then mix the two together. A sub-harmonic synthesizer would do the job also.

When editing audio, I never suffer clipping as I firstly create plenty of head-room. In the final mix, 0dB is my target, though my ear does not hear any distortion at +1.0 dB. In fact, I cannot distinguish between the two. But I can hear the difference when mixing orchestral tracks; for example, amplifying a bass instrument by even 0.5 dB can leave me feeling the vocal is starting to get buried.

Another thing I notice about the movie industry is that it is very presumptive by creating so much head-room in audio. I personally don't own a million watt (RMS) amplifier. OK. I'm exaggerating. Having worked as a recording artist at some good studios, like RCA, I learned that a "hot" signal greatly reduces noise inherent whenever amplifying a weak audio signal. The -4.6 dB of head-room in Popeye is torquephobia (yeah, I made that up).

Now since it might be too much to expect but never too much to ask, one more question: If -3dB lowers an audio signal by 50%, by how much in terms of percentage would the same signal be lowered if by -6dB? I couldn't find a chart that I could understand for the dB/% equivalents on the Internet.

[Cornucopia]

@Djard, headroom is a smart precautionary "best practice". And with the great majority of movies now being mastered in 24/96 (or similar), it makes sense to use substantial headroom. And with 24bits, you can "afford" it. If your total dynamic range was 144, but you put in 24bB of headroom, you still have 120dB. Lose another 20dB through processing (with dither), mixing, and other devices which might have less total DR, and you STILL end up with 100dB. That's still more that theoretically possible full DR with 16bit.

Using the amplitude terms hello_hello and pandy were so strongly referring to, -3dB would be ~70%, and -6dB would be 50%.

Scott

[hello_hello]

I dragged and dropped the movie Popeye (1980) into Audacity and juxtaposed the six tracks. No question in my mind that the LFE track has some unique frequencies. I am even able to create them by taking a sample from CH-1 and using the EQ to drop the pitch by an octave, and then mix the two together. A sub-harmonic synthesizer would do the job also.

I'll concede it's possible, but it's still a long way from the LFE channel only containing addition "effects" not found in the other channels.

When editing audio, I never suffer clipping as I firstly create plenty of head-room. In the final mix, 0dB is my target, though my ear does not hear any distortion at +1.0 dB. In fact, I cannot distinguish between the two. But I can hear the difference when mixing orchestral tracks; for example, amplifying a bass instrument by even 0.5 dB can leave me feeling the vocal is starting to get buried.

How do you determine in advance if Audacity exporting multi-channel audio is going to result in clipping? Maybe there's a way. I don't use Audacity much myself.

Now since it might be too much to expect but never too much to ask, one more question: If -3dB lowers an audio signal by 50%, by how much in terms of percentage would the same signal be lowered if by -6dB? I couldn't find a chart that I could understand for the dB/% equivalents on the Internet

The formula for converting amplitude ratio to dB is this (where -3dB is 70.7%, not 50%). It's the formula normally used for downmixing.

20 * log10(x)

20 * log10(0.5) = -6.02dB

50% = -6.02dB

I can't remember if you said you use foobar2000, but if you do and you run a ReplayGain scan you'll know the level it reports as the track peak comes in an obscure form such as 0.986745
The level it reports is an amplitude ratio which is converted to dB using the above formula.

There's a picture here displaying different dB values converted to either an amplitude ratio (formula above) or a power ratio.
http://en.wikipedia.org/wiki/Decibel

The formula you actually asked for, converting power ratio to dB, is (assuming my brain is working):

10 * log10(x)

10 * log10(0.5) = -3.01dB

50% = -3.01dB

And something I didn't know, which was written at the bottom of the image on Wikipedia.

The amplitude ratio is the square root of the power ratio.

200% is +3.01dB as a power ratio. The square root of 2 is 1.41421.
So if +3dB is 200% (power ratio) then it's about 141% (amplitude ratio).

Another thing I notice about the movie industry is that it is very presumptive by creating so much head-room in audio. I personally don't own a million watt (RMS) amplifier. OK. I'm exaggerating. Having worked as a recording artist at some good studios, like RCA, I learned that a "hot" signal greatly reduces noise inherent whenever amplifying a weak audio signal. The -4.6 dB of head-room in Popeye is torquephobia (yeah, I made that up).

Movie audio tends to be a lot more dynamic than most "music" audio, so it needs the headroom. And because it's also often multichannel, I guess that means it also requires enough headroom for when it's downmixed to stereo on playback. -4.6dB isn't a huge amount.
I think part of the issue with AC3 can be it's dialogue normalisation, which is designed to adjust the volume of AC3 audio (the entire audio, not just the dialog) so that dialog is always the same level. The idea being you can go from Forest Gump to (insert name of crappy super-hero movie here) and the dialogue will be the same level even if the dynamic range is different. This invariably seems to result in players which support AC3 dialogue normalistion reducing the volume to some degree. I don't know much about it or whether it can be disabled in most standalone players. Most PC decoders seem to ignore it.
Also, don't forget, much AC3 comes with dynamic range compression information which can usually be enabled in the player/receiver.

How's this for a co-incidence? I was looking for something fairly unrelated today and stumbled across this program.



As you can imagine my eyes initially did a Terminator-like scan looking for the channel assignment order (guess I need to research audio encode mode 7), but the dialogue normalisation stuff is there. This could be completely wrong, so take it with a grain of salt, but I think the standard dialogue normalisation level is -31dB, which means a player should decrease the volume of the AC3 audio above by 4dB, but as I said, that could be a load of crap. It's just the way I remember it. In an earlier post, Cornucopia linked to a pdf for Dolby encoding guidelines which probably explains how it works.

Here's some DVD AC3 audio:



The RF atenuattion value, I don't quite understand. Pandy mentioned it earlier. Maybe he'll pop back to explain. Likewise I don't really understand the "RF Ov. Pr. min/max" or "Dyn. Range min/max" it displays. If you go through the Dolby pdf and work it out.....

When you put the program into "tell me lots of stuff" mode, it offers lots of info for each frame I don't understand..... yet..... I'll probably end up researching it more myself now.

Unfortunately it isn't all that informative when it comes to AAC. It does however provide more information for DTS, I discovered after extracting some from a movie and then performing a Terminator-like scan of the results for the DTS channel assignments, which appear to be the same channel assignments listed on the Avisynth page I linked to originally. I'll have to research "audio channel arrangement 9" just to be sure. I don't want to practice my "I was right" dance for Cornucopia prematurely as there's an "I told you so" move in the middle with a fair degree of difficulty.

[hello_hello]

I assumed the centre and surround mix levels for the AC3 audio shown in my previous post were for stereo downmixing. It appears though, they're the levels for decoding the audio as 5.1ch surround. The stereo downmix levels are something else, assuming they're included. In this case they happen to be the same as the 5.1ch levels, which I don't quite understand, but anyway.....


[Attachment 27432 - Click to enlarge]

After converting some DTS to AC3 with AFTEN, the centre and surround levels are still set to -3dB. That must be the default for AC3. It must be to do with the way the surround system is expected to be calibrated, or something.....
I assume I remembered correctly when I said -31dB is the target dialogue normalisation level, so AFTEN would have set the level to -31dB and therefore the player won't adjust the level any further. I'm pretty sure that's how it works.


[Attachment 27433 - Click to enlarge]

[hello_hello]

DTSParser displays my DTS file as having two centre channels. I'm not sure why. That aside, it seems fairly keen on using the DTS channel order mentioned here.

[El Heggunte]

DTSParser displays my DTS file as having two centre channels. I'm not sure why. That aside, it seems fairly keen on using the DTS channel order mentioned here.

Me thinks DTSParser is incredibly outdated (though still somewhat useful, granted),
its author wrote it and left it behind before the rise of the Blu-Ray era
I mean, the program was not designed to deal correctly with DCA streams that are not compliant with the DVD-Video specification.

==========================================================
File ........: D:\tmp\test 4\321.dts
Size ........: 1886250 bytes

----------------------------------------- First Frame Info
CRC present .................: 0 (Not)
Number of PCM Sample Blocks .: 15 ( 512 samples/frame)
Primary Frame Byte Size .....: 1005 ( 1006 bytes/frame)
Audio Channel Arrangement ...: 9 (5 C + L + R + SL + SR)
Core Audio Samp. Frequency ..: 13 (48 kHz)
Transmission Bit Rate .......: 15 (768 Kb/s)
Embedded Down Mix Enabled ...: 0 (Not)
Embedded Dynamic Range Flag .: 0 (Not)
Embedded Time Stamp Flag ....: 0 (Not)
Auxiliary Data Flag .........: 0 (Not)
Mastered in HDCD format .....: 0 (Not)
Extension Audio Descr. Flag .: 0 (Channel Extension XCh)
Extended Coding Flag ........: 0 (Not)
Audio Sync Word Insert. Flag : 1 (Sub-sub-frame)
Low Frequency Effects Flag ..: 2 (Present, interpolation factor 64)
Predictor History Flag Switch: 1 (Yes)
Multirate Interpolator Switch: 0 (Non-perfect Reconstruction)
Encoder Software Revision ...: 7 (Current)
Copy History ................: 1 (Definition deliberately omitted)
Source PCM Resolution .......: 6 (24 bits)
Front Sum/Difference Flag ...: 0 (Not)
Surrounds Sum/Difference Flag: 0 (Not)
Dialog Normalization Param. .: - 0 dB
------------------------------------------- Estimated Info
Total Frames ......: 1875
Duration ..........: 20 seconds. ( 0 h. 0 m. 20 s.)
------------------------------------------------- End Info

P.S.: Also,

http://forum.doom9.org/showthread.php?t=170061

[Djard]

I'll try Foobar (yet another addition to my arsenal).

Examining other movie SS tracks, my impression is that movie sound engineers simply duplicate select parts of CH-1 and/or CH-2 then run the parts through EQ for the LFE channel, seeking to create as much separation as possible, which in the music recording industry is desirable and considered an art few have mastered.

When I import audio into the free but excellent Audacity studio, as far as I can tell, clipping is impossible; for the peak amplitude is never changed. Zooming into any series of "leveled" spikes to me indicates clipping that in my experience occurs only when someone has ripped a movie and applied their ignorance to the audio. I wrote a guide for the software (http://audacity101.blogspot.com/).

Am I understanding correctly the intimation that an encoded movie audio stream with peak amplitude boosted to 0dB can be distorted by a stand-alone amplifier for TV because it has no head-room? Surely that's impossible?

LeeAudiBi... yum! It offers info not inclided in another app I use often, called MediaInfo.

Thanks for the formula to calculate the ratio. I'll invest myself in that as I could use a little more dendritic sprouting in the left hemisphere of my organ of consciousness.

Most of what you gurus are citing is above my head, but Ill study it. Already I've learned that hello-hello is either from England, Ireland, Scotland, Australia, New Zealand or South Africa. How's that for a calculation?

[hello_hello]

Examining other movie SS tracks, my impression is that movie sound engineers simply duplicate select parts of CH-1 and/or CH-2 then run the parts through EQ for the LFE channel, seeking to create as much separation as possible, which in the music recording industry is desirable and considered an art few have mastered.

That's been my experience. The LFE channel mostly contains low frequencies already found in the other channels, just duplicated in the LFE channel for extra boost.

When I import audio into the free but excellent Audacity studio, as far as I can tell, clipping is impossible; for the peak amplitude is never changed.

We may not be on the same page...... I was under the impression the only option Audacity has for exporting multichannel audio was to downmix it to stereo (I don't use it much) but I just discovered it can export as multichannel. It even lets you remap the channels on the way through, although normally you wouldn't need to. If you're exporting it as multichannel then you should be able to import it, do whatever you like, then export it without having to worry about clipping.
If you're exporting as stereo though, the peaks can be clipped when the tracks are combined. If you take two audio tracks and combine them into a single track, the single track will be louder than either of the original tracks (individually). The same principle applies to combining multi-channel audio.

There's reasons why technically peaks of 0dB can still be clipped a little when importing lossy audio (due to slight changes in amplitude when audio is converted to a lossy form and back again, ie decoded), which is probably one reason why peaks of -3dB are often recommended. Any clipping would be pretty minor though.

Am I understanding correctly the intimation that an encoded movie audio stream with peak amplitude boosted to 0dB can be distorted by a stand-alone amplifier for TV because it has no head-room? Surely that's impossible?

If it's mutlichannel audio being downmixed to stereo, I'm not sure why it'd be impossible unless the downmix involves reducing the over-all volume to prevent the possibility of clipping. When standalone players are downmixing, they probably do just that.

Here's a couple of samples if you want to have a play (I pinched them from a post in another forum). The peaks in the "plus 4" sample are higher than 0dB. Here's the result of a ReplayGain scan (the wave file is 32 bit float so it can store values above 0dB, not the usual 16 bit integer which can't).



Using the formula for converting amplitude ratio to dB, the peak is +4.26dB.
You can import that file into Audacity and probably hear the peaks clipping on playback but they're not clipped in the sample. You could export that with Audacity as a 32 bit float wave file and the peaks would remain intact. You can also export it to some lossy formats (ie Ogg) without clipping the peaks as lossy formats can store values above 0dB. Whether the peaks are clipped mostly depends on the input formats they accept.

Take that file and simply export it as a 16 bit wave file, which can't store peaks above 0dB and you get this (clipped peaks):



Anyway.... the above isn't normal (peaks of +4dB) and something you'd want to avoid. It's just an example of what's possible.

The "zero" sample is the same audio in a 16 bit wave file with the peaks at 0dB (not clipped). Import it into Audacity three times so you have six channels. Export that as a stereo 16 bit wave file. I think you'll hear the clipping in the peaks when you play the exported version. That's what I'm referring to when I'm talking about downmixing multichannel audio and preventing clipping. If you reduced the volume of each track first (I think 6dB would do it) you should be able to export it again without clipping.
If you're not downmixing to stereo with Audacity that way, you can ignore everything I've said.

[hello_hello]

Anyone know how long Audacity has had an option to export multichannel audio? Or has it always been able to do so?
I remembered it only being able to export as stereo, but maybe I've just been remembering wrong.....

[Djard]

I've been using Audacity since version 1.2.6, and don't recall it ever being limited to stereo downmix. I'm sure folks at the Audacity forum will be able to answer that question.

I duplicated the posted audio sample six times in Audacity then downmixed it to stereo. Yep, the distortion is unacceptable. Thank you sooooo much for pointing this out to me.

Now I'm wondering why I didn't get any distortion when downmixing 52 tracks to a 4 channel mix, though I set the gain to 0dB on only a couple of tracks (see attached map). Maybe the wave amplitude is augmented only when identical frequencies are multiplied. No?

[pandy]

Guys 0dBFS can clip after digital to analog conversion (or oversampling and filtering - typical operation in sample rate converters).
http://www.indexcom.com/tech/0dBFS+/
https://forum.videohelp.com/threads/365050-AAC-Gain-Normalizing-to-0dB-Muxing-with-FFmp...=1#post2336836
https://forum.videohelp.com/threads/365050-AAC-Gain-Normalizing-to-0dB-Muxing-with-FFmp...=1#post2336896

http://www.audioholics.com/audio-technologies/issues-with-0dbfs-levels-on-digital-audi...ayback-systems

We can't ignore phase... as it will be not ignored in analog domain.

Adding 52 tracks need to be done in sufficient bitdepth (24 at worse, 32 or higher better) and all tracks need to be with reduced level - at final stage they can be normalized to -3.0103dBFS .

[hello_hello]

I've been using Audacity since version 1.2.6, and don't recall it ever being limited to stereo downmix. I'm sure folks at the Audacity forum will be able to answer that question.

Maybe I've just not paid enough attention, but Audacity has always offered a warning about audio being downmixed to stereo. Of course it doesn't when you select the multi-channel export option in it's settings, but I don't recall that option always being there.
The current warning says "your track will be downmixed to stereo", whereas I recall previously it said exported audio could only be stereo and to retain the multichannel configuration the "save project" option needed to be used. Or something like that. Maybe I'm just remembering it all wrong.

I duplicated the posted audio sample six times in Audacity then downmixed it to stereo. Yep, the distortion is unacceptable. Thank you sooooo much for pointing this out to me.

Now I'm wondering why I didn't get any distortion when downmixing 52 tracks to a 4 channel mix, though I set the gain to 0dB on only a couple of tracks (see attached map). Maybe the wave amplitude is augmented only when identical frequencies are multiplied. No?

Yeah you may have just been lucky due to the relative levels you'd set for each track. I'm not sure about the frequency augmenting part.
I kind of remember that if you play multiple tracks together and the stereo output meters don't indicate clipping, you should be safe to export to stereo too. I think you'll find if you imported the sample three times and adjusted the track volumes until you could play all three together without the output meters showing clipping..... well I'm sure you get the idea.

Edit 1: As pandy said in his previous post, -3dB is probably a better idea when converting to a lossy format, although I'll confess I've not been that fussy myself in the past.

The trouble is, playing the audio all the way through like that to find clipping wouldn't always be practical. You'd think there'd be a way to automate the mixdown level to compensate. Or check it before exporting. Maybe there is, and I just haven't found it yet. As I said, I don't use Audacity all that much.

I just discovered ffdshow's "normalise" downmix formula changes according to how you downmix, and it shows the values it's using as an amplitude ratio. I converted them to dB because ratios don't mean anything to me (as accurately as I could as it's sometimes hard to read).

When ffdshow downmixes 5.1ch to stereo with the LFE channel included (-3dB relative to left and right) and the normalise matrix enabled:
Left and Right -10.66dB, Centre -13.68dB, Surround, -10.66dB, LFE -13.68dB

If you follow the Dolby recommended mixdown (I think it's no LFE and -3dB relative to left and right for the surround):
Left and Right -7.64dB, Centre -10.63dB, Surround, -10.63dB

Add the LFE channel back in at -3dB relative to left and right:
Left and Right -9.70dB, Centre -12.70dB, Surround, -12.70dB, LFE -12.70dB

I'd never noticed it works like that before. You learn something every day. I guess they're all "worse case scenario" levels when it comes to preventing clipping. I recall the Dolby recommendation being to apply a 7.5dB gain reduction when downmixing, which makes sense looking at the above formula.

I've been in the habit of downmixing while applying a -6dB gain reduction and leaving it at that. It's always seemed to be enough the times I've checked..... but as a quick experiment I extracted some DTS from an action movie and downmixed it to stereo with foobar2000 (no further gain reduction). I used foobar2000's standard downmix DSP (I'm not sure exactly what downmix formula it uses). I converted it to AAC with Nero so any peaks above 0dB should be retained. I then ran a ReplayGain scan on the stereo file. The resulting track peak was +7.07dB. My usual -6dB gain reduction wouldn't have been quite enough. Damn.... I might have to re-think my entire outlook on life..... except, I hardly downmix to stereo any more. I wonder how I was checking it all originally when I came to the "-6dB is enough" conclusion? Maybe my method was flawed at the time. I can't quite remember, but damn!!!.....

Edit 2: As an experiment regarding what pandy said about -3dB being the better maximum level, I just downmixed the same file again while converting it to MP3. I'd originally thought LAME wouldn't clip the peaks, but checking again indicated it's maximum input bitdepth is 24 bit integer, when means any peaks above 0dB would be clipped. So I downmixed again, encoded with LAME and ran a ReplayGain scan. It reported the track peak as 2.85dB. So I assume those clipped 0dB peaks were being decoded at +2.85dB. I guess an extra 3dB of headroom is a good idea.

Edit 3: I just tested the downmix of another DTS file from a movie. This time the resulting track peak was +3.83dB, so my usual -6dB gain reduction when downmixing would have been enough. Maybe the first file I tested was a bit "untypical", although it did prove -6dB isn't necessarily always enough.

[Djard]

Trying to digest the wealth of info shared and weighed here, I feel like a sophomore among Ph.Ds.

From now on I'm going to give at least -6dB of head room to the loudest track--depending on the appearance of the waveforms--then, if the mix-down leaves head-room, I'll add some gain, prolly (sic) stop at -3dB. I'll let my ear through a good set of headphones be the final judge, auditing only the high amplitude sections that usually number only three or four in a 90 min movie; so the procedure will not cost much time. Of course if I amplify one track, I'll need to edit the others equally to preserve the original difference between the tracks. The -3dB head-room will still be a huge improvement from the -15 dB and > that seems common in HD BR movies.

Use of Audacity to assess for clipping is still appealing to me: after the mix-down, I can drop the movie (even VOBs) into the main panel and examine the waveform of the audio streams. Apart from being able to zoom in and actually see if the spikes have exceeded the 0dB threshold, I can click in the control panel of a track (highlighting the entire waveform) then click on Effect -> Amplify. The Amplify tool always reads how much head-room is available.

[hello_hello]

Use of Audacity to assess for clipping is still appealing to me: after the mix-down, I can drop the movie (even VOBs) into the main panel and examine the waveform of the audio streams. Apart from being able to zoom in and actually see if the spikes have exceeded the 0dB threshold, I can click in the control panel of a track (highlighting the entire waveform) then click on Effect -> Amplify. The Amplify tool always reads how much head-room is available.

That seems reasonable to me.
I'm still pretty sure though, if you play the audio after setting your relative track levels, and Audacity's left/right output meters don't show red (clipping), then it should be safe to downmix to stereo. I think the output meters retain the peak levels for a period, or even until you stop playing the audio, so it should be easy to see if there was clipping. You could probably check those "high amplitude" sections that way and not have to rely on your ears at all.

Aiming for peaks of -3dB is probably ideal, but I'll confess my enthusiasm hasn't stretched that far.

Edit: I did experiment with a few CD tracks. I checked the peak levels via Audacity's amplify function, converted the audio to MP3s, converted them back to wave (32 bit float), then checked the peak levels again. The second wave files had peaks between +0.4dB and +0.7dB higher than the first. Not a lot, but higher just the same.
The one exception was a track from a CD which appeared to already be clipped a little (looking at the waveform). After converting it to MP3 and back the peak increased from 0dB to +1.5dB.

I was wondering..... and maybe someone else will know..... my soundcard is configured to output stereo. Normally I downmix using ffdshow or the media player etc, but.....

When I was playing around with ffdshow's mixer I noticed that if I enabled it without checking the normalise option, the output level seemed to be the same as when ffdshow's mixer filter wasn't enabled. In other words, my soundcard seems to downmix multichannel audio to stereo without applying any additional volume reduction to prevent clipping. When I enable ffdshow's mixer and check the normalise option, the volume is much lower.
The logical conclusion seems to be, there's some degree of headroom above 0dB in the playback chain, at least when the soundcard is downmixing. Anyone know if that's correct, and if so, how much "headroom" would be typical? Not just for the soundcard, but for any PC speakers/amp connected to it etc.

Or if I play 5.1ch audio which the soundcard is downmixing to stereo and the result is peaks above 0dB, is it likely they're simply clipped?
Logically, or at least in a perfect world, the soundcard should be able to accept 0dB on every channel, combine them to stereo, output the stereo audio without clipping it (even if the output has peaks well over 0dB) and the connected amplifier/speakers should be able to reproduce it without clipping. In the real world though, I don't know if that's how it works.

[hello_hello]

I deleted most of this post as I tried an experiment re soundcard headroom but my testing was flawed, so the original post might confuse the issue.

The upshot is.... I played the sample I posted earlier with peaks at around +4dB using foobar2000, with foobar2000's output set to 24 bit. I could hear it distort (regardless of foobar200's volume control). When I changed foobar2000's output to 32 bit float I could stop the distortion simply by lowering foobar2000's volume control a little, but with the volume at maximum the audio was distorted in a similar manner.
My tentative conclusion would be the audio is always clipped if the output is 24 bit, but if it's 32 bit float it's not unless it exceeds the input the soundcard can handle. However as the 32 bit float output sounded like it was clipping in a similar way with the volume at maximum, it would indicate the soundcard doesn't have much headroom above 0dB on the input side, if any.

That'd be my conclusion.... which doesn't mean it correct.......

[Djard]

You've given me a lot about which to think.

Another way to create ample head-room is to manually attenuate the individual peaks of the few brief spikes in the waveform upon which the head-room is calculated by software and sound cards, providing the attenuation does not sacrifice too much of an important effect in the movie. For example the sound of a gunshot or door-slam lowered by -6dB--if tolerable--may permit that much amplification of the remainder.

The more I contend with the problem of what I regard as excessive dynamic range, the more I lean toward upgrading my home theater sound system, one that has control over dynamic range; but then I want the new iPhone 6 and am too frugal to buy both. But I might be willing to invest in Adobe's Premiere video editor if it can address all the issues one faces when contending with AV anomalies. As far as I know, Adobe does not support Matroska; and I don't see in the features list many of the functions I find myself needing to use when editing video files. And I think I now have about 40...each doing something another can't do. Have you tried Adobe Premiere?

[hello_hello]

I've never used Adobe Premier. My editing needs are generally fairly simple if I edit when I re-encode. So for me Avisynth has done the job (I use it with MeGUI). You can do a reasonable amount of "complex" editing with Avisynth, but it can get a hard to do, being script based. What sort of functions do you need when editing?

I'd still recommend trying some compression when you re-encode the audio, but maybe try making it easier than doing it with Audacity.
I don't know if you ever listened to the samples here, but you can easily tell the difference in dynamic range. The first half is just dialogue, the second half gunshots etc.
Originally I was applying that compression via DirectShow decoding and ffdshow, but I'd no longer need to (although I still compress the audio on playback that way rather than compress when encoding it). Since discovering the foobar2000 Winamp DSP Bridge and the Matrix Mixer DSP though, I can create a foobar2000 conversion preset which does the whole lot in one go without requiring ffdshow or DirectShow.

The Matrix Mixer DSP downmixes to stereo, normalising the way ffdshow does it when downmixing so you don't need to worry about clipping, the Winamp DSP Bridge loads the Winamp RockSteady plugin which compresses the stereo audio and then it's re-encoded. I use these settings, although you could always increase the "compression".
The whole lot is saved as a conversion preset, so it's just a matter of opening the 5.1ch audio, right clicking, and selecting convert......

Foobar2000 can open AVIs, MP4s and MKVs directly and play/convert the audio within. No need to even extract it first. It might take a bit to get your head around setting up foobar2000, but once you do the process should be quite fast. Well.... 1 minute, 50 seconds to downmix and compress an hours worth of 5.1ch AC3 and convert it to AAC using this old PC.

I compress audio the same way on playback and my PC speakers do the job fine (they're good quality for PC speakers but nothing out of the ordinary for hi-fi). I'm not constantly turning the volume up and down at night.

I should add.... I call the RockSteady DSP a "compressor" when it's not, because the end result is theoretically the same. Compressors squish down the loud bits then increase the overall volume. The RockSteady plugin works by only increasing the volume. The lower the level, the more it's amplified (depending on your settings). Unlike a compressor, which needs to be adjusted according to the input level to work effectively, the "amplify" approach can be far more "one size fits all". I found the settings which work for me and never change them.

[Djard]

I listened to the four files and found either of the DirectShow mixes waaaaaay better than the MeGUI Downmix that is typical of the dynamic range in BluRay movies. The Levelator mix utterly defeats the relationship between expected quiet and loud scenes. Too little dynamic range is equally undesirable.

Is RockSteady (executable file) solely for WinAmp or can the plugin be used with other apps, like Audacity? No kidding, but I think I have more than 40 apps and 100 plugins already for editing audio and video; yet I still find my arsenal inadequate...there's always one more thing required to fix some problem.

Most of what I do is demux video -> reduce excessive dynamic range and, if present, correct AV asynchrony (skew or drift. I rarely bother with asynchrony that alternates because it's too time-consuming, though Audacity can address that problem) -> and sometimes concatenate video that often causes AV asynchrony if using VDub, which usually requires -250 ms of correction of audio delay. Avidemux seems to do a better job of concatenation and muxing but has no audio correction feature. Then I use ConvertXtoDVD, so I can correct AV asynchrony if present and add menus for compilations. I use Aegisub to edit the almost always poor grammar of subtitles and use DVD Shrink to achieve file target size (4.38 GB); otherwise I use AVItoDVD. I like MP3Tag for album art and genre, etc; Audacity for editing and converting audio streams; BlueFab to back up my classic movies; Fairstars CD Ripper for CD rips; MediaInfo, Gspot and LeeAudiBi for media info that combined are still inadequate as none give the GOP structure; MPC HC w/K-Lite for auditing my work; VideoReDo to remove TV commercials; XMediaRecode or AVS Converter for recoding, correcting AR, etc.; and my old favorite VirtualDub for trimming, etc.; I also use IFOEdit, AR & bitrate calculator, autoruns, Avisynth, screen capture, Chopper (VOBs), Disk Cover Creator, PowerISO, IMGBurn, MKVToolnix, MP4Joiner, Ripit4Me (when updated BlueFab fails), and many others.

So I wonder if a program like Adobe Premiere could reduce the need to use so many apps?

[Detmek]

@hello_hello
If you use foobar you also have R128Norm - a DSP compressor/normalizer based on gated ITU-R BS.1770 Loudness calculation. It acts like compressor/normalizer but normalizes everything to -89dB, which gives a lot of headroom. But, just in case I would add Advanced Limiter to prevent clipping.

[hello_hello]

I've seen the DSP, but never really understood how it works, and I'd not got around to trying to find out.
Although I vaguely recall reading this post and deciding if it was accurate, it wouldn't be all that great. However I ran a test encode (see later in this post).

The 89dB thing...... it's hard to explain..... if I even remember correctly...... but the ReplayGain target volume is based on a SMPTE standard for translating what you see in your audio editor into sound pressure level (using the dB scale) when a movie theatre system is properly calibrated. ie if you run pink noise through this theatre system at this voltage then this is how loud it should be, 89dB... sound pressure level. (except I think the SMPTE standard is actually 83dB to allow for more dynamic range before clipping). It's not actually "normalising to -89dB" as such. I couldn't tell you how to convert that 89dB of sound pressure level back to a dB level in your audio editor, if it's possible. I assume it is, but I'd need to research that one. The whole thing does my head in a little. Maybe you'll have better luck.
http://wiki.hydrogenaud.io/index.php?title=ReplayGain_1.0_specification#Reference_level

ReplayGain works by analysing audio and trying to determine how loud it sounds. The standard ReplayGain target "volume" is 89dB. I adjust all my music with ReplayGain before transferring it to my MP3 player as that way everything sounds very close to the same in volume. I've used MP3Gain for years, although you can now do the same with foobar2000. The main difference is MP3Gain saves info to the MP3 tags to make the process reversible (you can adjust the volume of MP3s losslessly). foobar2000 doesn't.

All ReplayGain normally does is adjust the volume of each track as a whole. It doesn't compress, so the dynamic range doesn't change. It appears the DSP to which you referred analyses the audio in small sections and adjusts the volume up and down to try to keep it close to the target volume of 89dB. In theory I didn't think I'd like it, but.....

I took the MeGUI downmix sample from the other thread and converted it twice. Once using the R128Norm DSP and then again using my standard RockSteady settings. I then ran a ReplayGain scan on the encodes. The encode with the R128Norm DSP was only 0.17dB below the 89dB target volume, so it got very close. The peaks were about +3db so I converted it again with the advanced limiter too.
To compare that to the RockSteady version I ran a ReplayGain scan on the RockSteady conversion and adjusted the over-all level to the same 89dB target volume. Once it was adjusted to 89dB it's peak was 3.81dB below maximum, so I guess the RockSteady plugin compressed just a little more.... on average.

The R128Norm DSP surprised me, in that it did a much better job than I expected. There wasn't a lot of difference between it and RockSteady. If anything, the dialogue in the first half is a tad louder. The main thing which had me declaring RockSteady the winner was the very first gunshot. It's louder in the R128Norm version. Not the end of the world though. I re-encoded each as AAC and attached them if you want to listen.

Rocksteady is just a Winamp plugin although I use it with foobar2000 thanks to the foobar2000 Winamp DSP.
You certainly use a large number of programs. Me...... MeGUI, foobar2000 are mainly what I use, along with MKVMergeGUI and MP3Gain, but if it's any consolation, I have a "multimedia" folder in my start menu, and the contents of it look like this:



Fortunately about the only time I create a compliant video DVD these days is if I need to give something to someone with only a DVD player to play it, so although I have AVStoDVD I rarely use it.

[Detmek]

OK, I will check samples when I get home.
I mentioned R128Norm because it is a native DSP plugin for foobar. In the past I had problems using Winamp DSP Bridge:

1. Using anything but 16-bit internal processing in most cases caused crash and I would like to avoid processing in 16-bit,

2. Didn't verify this but Winamp API works in fixed point only and so Winamp DSP Bridge. Foobars internal processing is 32-bit float so audio must be converted to fixed point and then back to floating point when WDB is in filter chain,

3. WDB has built-in hard limiter, -3dB or -6dB, I can't remember now but I prefer to put limiter as last filter in filter chain, not in the middle. OK, that limiter is useful when you work in fixed point as you can't recover audio from clipping if audio exceeds full scale during prosessing. But, I would like to avoid limiting audio berfore I finish processing.

P.S.

[hello_hello]

Detmek,
Sorry, I obviously didn't look closely enough and thought the two posts prior to my last one were both by Djard, so I replied to both in a single post. If you wondered what I was babbling on about at times......

The way I look at it, is if you're compressing audio for playback at night or for portable devices etc (which I do on playback), by comparison a 16bit bitdepth is probably fairly low on the "horrible things to do to the audio" scale.
16 bit is CD quality so it's not exactly terrible. If you're doing lots of processing then a higher bitdepth might be nice but I can't say I'd fuss about running a couple of DSPs over 16 bit audio. As a general rule, you'd probably put compression at the end of the DSP chain, so any DSPs prior to the Winamp bridge could still use a higher bitdepth.

Not all encoders accept a 32 bit float input anyway. Most of them are at least 24bit but the latest flavour of the month AAC encoder, FDKAAC, downsamples a 32bit/24 bit input to 16 bit before encoding.
I couldn't seem to cause any crashes by trying to run RockSteady in something other than 16 bit mode. When I did (via the Winamp DSP Bridge) all that'd happen was RockSteady would no longer work. It's the same when I run it via the ffdshow Winamp plugin. 16 bit only.

I wouldn't take the "quality" of those samples as being typical. I used the "MeGUI downmix" sample from here, which was AAC, converted it to a 16 bit wave file while applying each compression type, then converted each back to AAC again (while applying ReplayGain to the RockSteady version). Hopefully there's no noticeable artefacts from the double lossy encoding. It was really only done as a quick test to compare the R128Norm compressor DSP to the settings I normally use with RockSteady. I'll probably try some more experimenting with the R128Norm DSP at some stage as it's obviously pretty good. For compressing the audio on playback though (movie soundtracks, TV shows etc) RockSteady/ffdshow is still my only option. I can't use the R128Norm DSP for that.

[Detmek]

No problem.

I don't mind processing audio in 16-bit for playback but if I encode that same audio I prefer to do processing in 32-bit float, just in case. I might noticed artifacts latter when I watch movie or listen music. Encoding can be done in 24-bit, no problem with that.

I listen your samples. Both sound good to me. R128Norm is a bit loader and has more dynamics. Not a bad thing but I don't know how original sounds so I can judge result. But the end result is good. It also works on multichannel audio. Don't know if Rocksteady supports multichannel.

Rocksteady is more of "night mode" as lower dynamics makes it more pleasent to listen, well at night when you do not want to disturb your neighbors or family. And it is useful for real time playback.

I will have to experiment with both plugins on few movies to judge but normalizing at playback is probably better idea then to permanently alter audio track. I was thinking of adding 2 tracks to file, original and downmixed-normalized stereo.