NEW CODEC! 128 kbps and less for 5.1 surround encoding for MPEG4 (Xvid/Divx) movies [Archive] - Page 2

3dsnar

11th January 2006, 20:11

I am suggesting to use VirtualDubMod_Aud-X_enabled
for encoding films :)
because this is the most convenient way to combine it with Xvid/DivX movies.
For separate audio files, ofcourse th exe frontend is the best choice.
---
For those who want to utilize our exe in their own apps, here are the command line complete commands (for encoding from 6 channel PCM to Aud-X. Please note that the channel order is of importance)
---
audx.exe STRQ FileNameWithExtention
audx.exe STDQ FileNameWithExtention
audx.exe HGHQ FileNameWithExtention
audx.exe SPBQ FileNameWithExtention
---
DECODING / TRANSCODING TO AC3
audx.exe PCM AudxFileNameWithExtention
audx.exe A128 AudxFileNameWithExtention
audx.exe A192 AudxFileNameWithExtention
audx.exe A256 AudxFileNameWithExtention
audx.exe A384 AudxFileNameWithExtention
audx.exe A512 AudxFileNameWithExtention
audx.exe A640 AudxFileNameWithExtention

3dsnar

11th January 2006, 20:14

I will try to find out how to overcome the problem with filter registration
and will report the solution ASAP.
I am sorry for this bug:confused:

Rockaria

11th January 2006, 20:35

Thanks for the reply. I apparantly missed one crucial issue.:)
Does the audx.exe encoding support stdin(pipe) as the input wav source?
In other words, can I replace the 'FileNameWithExtention' with '-' ?

Based on my test, it does not seem to support the pipe(linked from many different audio format decoders & frame servers) yet.

3dsnar

11th January 2006, 20:43

unfortunately not. The file name is necessary.

Teegedeck

11th January 2006, 23:20

I say, the whitepaper reads a bit like the technology is not very different from MP3-encoded Dolby-Prologic, only with some additional channel-info as 'pointers' of sorts. Is that so? Or do you use actual spatial encoding the way it will be implemented in HE AAC?

If the surround info is hidden in a channel by phase-shifting before it is encoded to MP3 that gives me an uneasy feeling. You know that the psycho-acoustic component of MP3-encoders (You use LAME?) doesn't react well to such tricks? Simply because MP3 isn't exactly meant to preserve such non-standard information.

3dsnar

12th January 2006, 07:58

Thanks for your thoughts.
The signal is merged to stereo/mono (identical precedure is in HE AAC and mp3surround)
and encoded with LAME. We did not change anything in LAME, i.e. we did not manipulate in the psychacoustic model in the mp3 encoder. Before merging the channels, some information describing how to 'unmerge' the signal are separately computed. And this is the algorithmic solution that we developed. Mp3surround and HE-AAC has the same approach (i.e. providing information for 'unmerging' and algorithm to carry out such operation).
As far as I am concerned these two technologies are focused on variations of binaural cue coding.
---
We used a completely different approach got preserving the spatial info.
The difference is that we put emphasis on phase distortions of the restored 5.1 sound (it happens after 'unmerging' the mp3 main stream to PCM and has nothing to do with LAME code).
----
Phase is extremely important for movie sound tracks!
Therefore we have mostly tested and tuned our codec on movie sound tracks mainly because of one very important reason
(I started to play with multichannel music,
but I decided to focus on movies only).
Very often in movie sound tracks channels contain the same sound components, but of shifted phase (phase shifts vary for different frequencies).
For example in movies like "Fifth Element" nice ambient spatial feeling at the begining was achieved based on phase shifts of some spectral components.
PHASE REALLY MATTERS FOR THIS MASTERPIECE OF 5.1 SOUND,
and as well for other movies with good surround audio.
-----
This makes coding spatial information EXTREMLY difficult, because
merging channels ends up in loosing the phase relations.
So after reconstruction, even when you reproduce energy relations between channels, phase gets completely distorted (I mean the phase shifts of spectral components between channels are lost).
I tried to show it in my whitepaper when I did the objective test
(I know this is not an appropriate way to compare codecs and subjective tests are much better, but this was the only thing that we could produce at that time).
And it shows phase and energy erros of our codec, AC3 for comparable bitrates and mp3surround.
Mp3surround is really poor in preserving phase.
So when you watch movies like "Fifth Element" on high quality
5.1 home theatre system, really alot of the spatial feeling is gone...
And we did everything to avoid this!
----
Interestingly the phase problem is not so important
for multichannel music (in opposite to multichannel movie sound tracks),
because the channels mostly differ in energy (if they have some common components).
Also, in subjective tests ONLY this type of audio (i.e. 5.1 music)
was used for evaluation of the encoders.
(HE-AAC and mp3surround)
(check below links with whitepapers)
http://www.chiariglione.org/mpeg/working_documents/mpeg-d/sac/RM0-listening-tests.zip
http://www.iis.fraunhofer.de/amm/download/flyer/dl.html?f=introduction_to_mp3surround.pdf
===
To summarize, our codec is dedicated to movie 5.1 (surround) sound tracks.
It would be interesting to see a subjective comparison
of Aud-X, mp3surround and HE AAC...
Maybe someone objective (some organization specializing in such tests, not the codecs scientists)
will perfm it...
IMO subjective tests for movie sound tracks and 5.1 music should be carried out separately,
since both differ in terms of production (utilizing phase shifts for certain spatial effects is used only in movies)

Teegedeck

12th January 2006, 08:47

So the approach is very similar to actual BCC, and closely related to MP3-surround if I understood that correctly. :) AFAIK encoding phase differences also is part of BCC.

Does the encoded signal of aud-x of also carry information about the channels' level difference, time difference, and coherence?

3dsnar

12th January 2006, 09:12

The phase info must be very limited in mp3surround (please see our comparison tests in our whitepaper),
http://www.aud-x.com/images/stories/downloads/aud-x_whitepaper.pdf
because the reconstructed 5.1 sound has very distorted phase.
Since it was US who did perform the comparison (in our whitepaper), and we are not very objective here ;)
you can use our front end to decode Aud-X stream (and for example winamp
to decode mp3surround) and perform the tests on your own (listening on 5.1 speakers set and objective evaluation in Matlab, or something like that)
---
Yes, ofcourse, Aud-X carries energy information related to various frequencies. Energy based error (also presented in the whitepaper) is also
significantly lower, than in case of mp3surround.
It should be stressed that we used very difficult movie sound track fragment,
for the test (fragment of Luc Besson's "Fifth Element"),
and maybe for 5.1 music the comparison would result in similar performance...I do not know.
Our goal was to provide a solution dedicated to 5.1 audio in movies :D

dimzon

12th January 2006, 09:49

Thanks for your thoughts.
The signal is merged to stereo/mono (identical precedure is in HE AAC and mp3surround)
and encoded with LAME.
LAME is LGPL software! Does You pay for it? Otherwise, please provide source code!

3dsnar

12th January 2006, 09:58

We did not modify LAME...and as you noticed it is LGPL.
Aud-X code is something separate
(I hope I explained it clearly in the previous posts).
Our software is a freeware

dimzon

12th January 2006, 10:04

We did not modify LAME...and as you noticed it is LGPL.
Aud-X code is something separate
(I hope I explained it clearly in the previous posts).
Our software is a freeware

1) Does it mean you can use any other lossy encoder instead LAME? Maybe You can provide some SDK to be able to connect Your algorytm to any encoder (read: Ogg Vorbis).

2) Can You provide full algorytm/bitstream specification (follow RED LINE from my signature to understand why I want it)

3dsnar

12th January 2006, 10:09

These subjects were exploited in this thread.
Please refer to the previous posts.

dimzon

12th January 2006, 10:20

These subjects were exploited in this thread.
Please refer to the previous posts.
I have re-read all thread again and dont found Your answers on
Maybe You can provide some SDK to be able to connect Your algorytm to any encoder (read: Ogg Vorbis).
and
Can You provide full algorytm/bitstream specification
You wrote:
BTW. If anyone can help in any way
to popularize the software, that would be ofcourse great!
(among users, movie coders, etc)
Any ideas or suggestions?
And I'm trying to help You! I understand why does you chose mp3 as carrier bitstream but I really believe Ogg Vorbis - based impementation woud be a really revolution!

3dsnar

12th January 2006, 10:31

Yes you are right. Cobining our technology with Ogg Vorbis would be great!
We really want to try it and talk to guys from Vorbis (anyone here?)!
Honestly, I am really exhausted by this project,
thus working on the next stage will be possible after couple of months
(I need to take care of my wife and kids, and finish my PhD ya know ;) )
But seriously. We are exactly going to do as you suggest (if Vorbis group is interested). I will keep everyone informed!

shon3i

12th January 2006, 17:02

Did you find solution for ax file

3dsnar

12th January 2006, 17:13

I am sorry, not yet.
I am going to work on this over this weekend.
Also, after the weekend I will update the installer,
so now using Aud-X will not exclude using FFDshow.

shon3i

12th January 2006, 19:46

OK Thanks

shon3i

12th January 2006, 23:31

You ain't gonna believe this. I'm managed to start the codec by deleting it first, and then, after restarting the computer a couple of times, I've reinstalled it and it worked. Conclusion: I think HE-AAC is better than this one. Why does DirectShow Filter overrides mp3 stream?

kotrtim

13th January 2006, 05:22

well, instead of movie, I tried to ecode audio CD with Aud-X. Aud-X 128 kbps sounded worst than QT LC-AAC 80 kbps!!!!
Why don't just drop Aud-X project, no point keep using mp3, since AAC is made by MPEG to replace mp3
Vorbis is superior to mp3, Besides, I can't figure out how to decode the aud-X wav file with a mp3 decoder.....
It's better to just merge/joint force Aud-X (Free) project with Vorbis (OSS).
Create an enhanced Vorbis "Vorbis 2" which have SBR, PS, "fake surround", optimized channel coupling for surround....isn't that cool? We must discourage people from using avi, Vorbis doesn't work properly in avi...
Encourage people to use MKV instead......

Rockaria

13th January 2006, 06:05

I find no point of mentiong the quality on 2 ch transcoding except the (dynamic) 5.1ch upsampling (from 2ch CD) through the DS filter or aud-x ac3 encoding which any aac solutions do not have.

It is basically a technology pseudo embeding & reproducing the multi channel stream into an existing popular format and looks a lot enhanced than the original mp3 surround solution from FH.
Of cource, if the vorbis encoder can keep the auxilary spatial information better than the Lame encoder, the vorbis can be considered a better carrier.

I am listening an original DTS, aud-x shrinked mp3 and aud-x encoded ac3 5.1ch music. No noticible difference found so far compared to the AAC HE VBR except a bit lower volume level possibly because of the different DS Filter decodings. Maybe I need some spectral image comparison tests on before & after images. The only trouble is this machine is not setup for such tests yet.

vinouz

13th January 2006, 17:43

3dsnar:
I didn't see any post on HydrogenAudio.org ? It's the place to get heavily tested/commented/advised for sound encoding.

Seems impressive. I'd like to test your 'Superb Quality' settings with an ogg q4 mono at the core.

And your internet quality on an ogg q0 mono core too...

kotrtim:
As for testing 128kbps Aud-x vs 80kbps LC/AAC, I find the comparison useless : probably Aud-x is using mp3 around 80-96kbps at the core, and surely it is easily beaten by LC-AAC on that segment. No wonder.

3dsnar again:
So, interesting proposal, but : what amount of bitrate does your scheme add (In the four different modes. At min, average, and max.) ?

I'm glad a new codec is developped, and I'm also glad it seems you have solid grounds on which to continue. Finally the presentation, the restricted mode choice and the application domain seem well thought (IMHO). And I wish such an initiative a good luck.

If it appears promising, I might as well mention it in my A/V compression courses.

3dsnar

14th January 2006, 07:47

Thank you all for tests an all the opinions.
Please note. What we were trying to do, is to
work on preserving the spatial feeling of 5.1 sound. Thus if our core mp3
stream sounds worst than core AAC (due to better sound quality of AAC than LAME mp3) it is a different story, and beyond the scope of our work.
--
This scientific experiment was meant to preserve the phase relations in movie sound tracks for low bitrates.
Therefore it is important to test it with a 5.1 home theatre system (or 5.1 system set) against
mp3surround or HE-AAC.
And I think that users of such equipment will appreciate our results.
--
As I explained previously, the reason to chose mp3 as a carrier was that
we wanted the Aud-X stream to be compatible with AVI. AVI is crap...but very popular. That is all.
--
It is very important to remember, when using our executable front-end for encoding Aud-X, to feed the encoder with a proper channel order signal!
Please refer to the information in HELP.
--
Please look carefully at Hydrogenaudio, because
I have submited alot of posts there, and the discussion
was interesting :)

3dsnar

14th January 2006, 08:05

For those who tested Aud-X on music.
1) Encoding stereo music converted to 5.1 makes on sense for two reasons.
The decoder contains pseudosurround algorithm, thus when you play a
stereo track, by chosing 5.1 output or SPDiF (for external amplifier), the
algorithm will be applied, and there is no need to encode such tracks
with Aud-X.
------
2) It is not dedicated to 5.1 music. Please try to find a nice movie sound track, which sounds really good on your 5.1 sound system and perform the tests.
------
Making the comparison you a stereo system (with headphones), or listening to each stream with headphones makes no sense! As I mentioned, we focused on recreating phase relations of various channels, which is important only on 5.1 speaker set. And this is the only reasonable comparison environment.

3dsnar

14th January 2006, 08:12

For those who performed the tests:
Please provide links to the tested sound (so I will be able
to focus on potential problems for future releases)
1) Aud-X signal
2) Competitive signal of similar bitrate
3) Original signal (used for encoding).
-----
Thank you.

3dsnar

14th January 2006, 08:23

The core bitrates that we use
1) STRQ 80 kbps -> base mp3 is 56 kbps mono (24 kbps for spatial info)
2) STDQ 128 kbps -> base mp3 is 64 kbps mono (64 kbps for spatial info)
3) HGHQ 192 kbps -> base mp3 is 128 kbps stereo (64 kbps for spatial info)
4) SPBQ 192 kbps -> base mp3 is mono 64 kbps (128 kbps for spatial info)
---------------------------------------------------------------------
So from above you can see that in fact SPBQ has no better base sound quality than STDQ. We decided that 64 kbps LAME 3.96.1 is good enough, and the rest of kbps should be used to recreate the 5.1 panorama and spatial feeling. This and only this was an objective of all the research that we carried out. Improving base psychoacoustic encoder was beyond the scope of our work (any did not modify LAME at all)

The emphasis was put on the recreation of phase relations (i.e. spatial feeling). Please
check out our whitepaper to see how SPBQ performs (in terms of phase and energy error) with reference to AC3 192 kbps and mp3surround 192 kbps.
http://www.aud-x.com/images/stories/downloads/aud-x_whitepaper.pdf

tebasuna51

14th January 2006, 11:22

About HGHQ 192 kbps option.

Decoded with a mp3 decoder I get a stereo output but with two equal channel (FL = FR), then is not a real stereo output. If this is always true, I think the users must know this issue (important for standalone players).

Decoded with Aud-X decoder, is there any reason to use HGHQ instead SPBQ quality?

3dsnar

14th January 2006, 11:37

No, this should be a normal stereo (otherwise we would not make this option
available at all):
To make sure, I just tested it and it is 100% ok.
So:
1) Maybe the supported (for encoding) stream order was not correct?
2) Maybe by mistake you used SPBQ instead of HGHQ?

tebasuna51

14th January 2006, 13:36

Sorry, is only for a 'special' test input.

In other test I obtain a correct stereo output with:
FL' = 0.4xFL + 0.2xC + 0.4xSL
FR' = 0.4xFR + 0.2xC + 0.4xSR

3dsnar

14th January 2006, 16:15

If you are so deep in the test :)
please check how much better is:
1) front vs rear channels crosstalk for HGHQ than in case of mp3surround
2) front vs rear channels crosstalk for SPBQ than in case of mp3surround
(i.e. in Aud-X there is no front vs rear channels crosstalk for SPBQ)

Rockaria

17th January 2006, 15:59

While performing the listening tests prior to any possible professional ABX tests, I found some usability problems making it hard to compare to other format dsfilters play. Overriding the dsfilter merits and filter orders in the players(I tested with MPC, gomplayer, bsplayer and zplayer) will allow choosing the dsfilter for the (aud-x) mp3/ac3 play without (un)installations. Also the new installation of the aud-x dsfilter does not show the illegal memory address error on normal mp3 upmix play any more.

Below are the check points I've tried to figure out while reviewing & testing for this new promising FREE codec.

<aud-x dsfilter decoder + alpha>
mp3(normal) -> decode + upmix/48k resample + apu rendering/ac3 encode & spdif passthrough
mp3s(aud-x)-> decode + apu rendering/ac3 encode & spdif passthrough
ac3 -> (decode/resample) + apu rendering/(ac3 encode) & spdif passthrough

1. indeed a complete set but with no DRC or gain control? cannot connect to FFDShow or AC3Filter to utilize the DRC or other DSPs
-> provide the DRC/gain/EQ control or allow connecting to other filters, the volume level is critical in perceived sound quality for normal ears.
2. cannot be used for AVISynth DirectShowSource frame source, looks too strongly tied to the sound (out) device.
3. what kind of an algorithm is used for the mp3 upmix, martrix. DPL or DLP II? can users choose or set the parameters?
4. what else formats can it decode besides the mp3, mp3s(aud-x) & ac3?

<aud-x cli encoder>
1. restricted to the wav file sources. when the duration is too long, i.e. over 1hr, the temporary disk usage would be a big problem.
-> allow std-in stream for the wav source for tons of existing cli decoders for transcoding as I said before
2. why the bit rate is lmited to 128k &192k CBR? even the nero AAC HE VBR streaming quality is around 33% bigger than aud-x.
-> allow VBR for non-avi environment
-> rescale(upsize) the bitrates on the quality classes(internet, streaming...) in real quality grades, complicated signals require bigger size for the same perceived quality.
3. would the lossless(FLAC) or hybrid(wav pack) formats be better as a container for the base & spatial data? If the signal is compressed & extracted through the lossy format, any loss in the spatial(& base) data would be critical in the reconstruction of the channels. It would be interesting to see how much space can be saved by using the lossless format, if any, for multi channel encoding.

<aud-x acm codec for the specific vdubmod only?>
1. as is pointed out, it does not appear anywhere else except the vdubmod audio stream codec selection.
2. also with the help of many mux tools, the movie transcoding does not need to perform the a/v transcoding together, making the standalone encoder more usable.(it worries me repetedly emphasized for the movie soundtrack use..)

If set up in the identical decoding(playing) environment except the decode function of the dsfilter, it would be an easy & resonable job to perform the ear-comparison without any biases(volume, DRC, EQ and many hidden other factors).

Regards.

3dsnar

17th January 2006, 16:51

Thank you for all the really valuable suggestion.
We will work on future improvements.
-----
For the subjective tests, please use our executable front-end
(to decode the Aud-X stream to PCM).
-----
Cheers, 3d

Rockaria

18th January 2006, 04:38

I have some result to share as for the listening test performed on aud-x mp3 and aud-x decoded wav compared to original dts-decoded wav.
Initially, I tested on my digital system(AN7(soundstorm) connected to ONKYO TX-L5) and an analog headphone(5.1 Kinyo in stereo dolby surround encode mode), which resulted in fairly good quality considering the low bit rate of SPBQ(192kbps CBR).

But when I decided to use the Kinyo in 5.1ch mode today, I finally could find the crosstalks all across the channels, which I could manage to minimize by setting the 'invert phase' for rear and center/lfe channels. HGHQ was better than SPBQ in my case incrosstalk-wise.
But the digital is gone bad, the sound space has changed/reversed/flattened with the invert phases!

There was no crosstalks for the original wav. Maybe a special care for the analog surrounds phase handling is reuired.

Regards.

3dsnar

18th January 2006, 07:40

OK. Thanks for the tests!
We has still some things to do (such as incorporating Prologic II)
etc. So your suggestions are very valuable for us.

> There was no crosstalks for the original wav.
> Maybe a special care for the analog surrounds phase handling is reuired.
Indeed! This also requires additional work (i.e. sounds encoded with Prologic).
Therefore we mus incorporate this technology in Aud-X.
We hope to do it before the next release.
--

3dsnar

18th January 2006, 08:49

BTW. I forgot to ask.

What should be the library format to incorporate Aud-X with BeSweet (and BeLight)?

dimzon

18th January 2006, 11:41

BTW. I forgot to ask.

What should be the library format to incorporate Aud-X with BeSweet (and BeLight)?
For BeHappy integration I need command line console encoder wich can handle data from stdin

3dsnar

18th January 2006, 12:13

OK, thanx for info

Rockaria

25th January 2006, 14:48

ac3 -> (decode/resample) + apu rendering/(ac3 encode) & spdif passthrough
...
4. what else formats can it decode besides the mp3, mp3s(aud-x) & ac3?
..
But when I decided to use the Kinyo in 5.1ch mode today, I finally could find the crosstalks all across the channels, which I could manage to minimize by setting the 'invert phase' for rear and center/lfe channels. HGHQ was better than SPBQ in my case incrosstalk-wise.
But the digital is gone bad, the sound space has changed/reversed/flattened with the invert phases!
It does not (seem) to play the ac3 anymore(sort of a conflict between the filters?). Instead, I want to mention it can also play 5.1 wav.
Recently, just before the aud-x test,I changed my test machine from au13 to An7, both soundstorm, which seems to have made the DAC & analog response change on the phase. It sounds better now with the invert phase checked in the nvmixer (hidden) option regardless of the decoder/player with the analog speaker.
But the analog crosstalks only happened with the aud-x transcoded & played mp3/wav/ac3 when the phase invert are unchecked on this machine.

Also I want to mention the rear encode & decode routine of the DPL II would be useful for the reference.

3dsnar

25th January 2006, 16:13

Rockaria,
thanx for the test.
Yes. I have disabled AC3, since people were complaining that Aud-X
replaces AC3filter. So to avoid such a conflict, I've decided not to decode AC3.
--------
The phase experiments that you've performed are interesting. However they may depend on the signal (phase relations in current signal). In general there is a channel cross talk for certain Aud-X encoding options (for example for STDQ = 128 kbps). For SPBQ and HGHQ it is significantly limited.
But the most important thing is that phase relations between channels are preserved (i.e. the crosstalk concerns the spectrum elements, for which phase and spatial perception are not linked with each other).
And in general this is what Aud-X is all about -> preserving phase relations, which are responsible for the spatial feeling.
---------
It is also advised to use original AC3 (or PCM generated from AC3), rather than DPLII decodec multichannel signal, because the problem with DPLII
is that it forces specific phase relations in the surround panorama.
Thus when you convert DPLII to Aud-X, you will not gain increase of quality, but rather more crosstalk (the original DPLII phase distortion will be preserved, since Aud-X tries to keep the inputs phase).
---------
I do not know if you have noticed. But when you feed a stereo sound to our decoder,
and choose 5.1 output, pseudosurround effect is performed. I am curious of your
thoughts regarding this effect :)
Please listen to it in 5.1 speaker environment.

Rockaria

25th January 2006, 19:30

OK. Thanks for the tests!
We has still some things to do (such as incorporating Prologic II)
etc. ..
> There was no crosstalks for the original wav.
> Maybe a special care for the analog surrounds phase handling is reuired.
Indeed! This also requires additional work (i.e. sounds encoded with Prologic).
Therefore we mus incorporate this technology in Aud-X.
We hope to do it before the next release.
..
The phase experiments that you've performed are interesting. However they may depend on the signal (phase relations in current signal).
--
(5.1 Kinyo in stereo dolby surround encode mode)
Well, maybe you misunderstood my initial test about the crosstalks with the knyo 5.1ch analog headphone.
The soundstorm has a DSP in the nvmixer encoding/downmixing 5.1ch signal to 2ch DS/DPL for analog speakers/headphone. I tested the kinyo in this 2ch DPL mode then original 5.1 mode. The latter produced the crosstalks as I mentioned before and disappeared when I set the invert phase on center & rear.
The aud-x encoded then decoded 5.1ch wav generated the crosstalk when played with aud-x while none with other players/dsfilters. Also the digital connection has no crosstalks at all.
So it cannot be concluded it's a signal(encoded) problem alone.

Also now I am curious why you mentioned the DPL II when I reported the crosstalks between the channels. I thought that it( reference) would be only useful to eliminate the crosstalks in the signal(encoding) and decoding(aud-x) happening on certain DAC & analog speakers.

Thus when you convert DPLII to Aud-X, you will not gain increase of quality, but rather more crosstalk (the original DPLII phase distortion will be preserved, since Aud-X tries to keep the inputs phase).
...
And in general this is what Aud-X is all about -> preserving phase relations, which are responsible for the spatial feeling
I never think of myself transcoding the DPL II content to aud-x as well as mp3 to aac. The scenario itself already contains some big bias.
When the aud-x embed the surround data with the phase shift technology then seperates the channels with aud-x filter, how much channel seperation is reserved other than the abstract 'sprtial feeling'? Is there anything artificial sound generation related?

3dsnar

25th January 2006, 20:06

OK, indeed I misunderstood what you meant

> I never think of myself transcoding the DPL II content to aud-x as well as mp3
> to aac. The scenario itself already contains some big bias.
> When the aud-x embed the surround data with the phase shift technology
> then seperates the channels with aud-x filter, how much channel seperation is
> reserved other than the abstract 'sprtial feeling'? Is there anything artificial
> sound generation related?

I am not sure if I understand your question correctly.
In general Aud-X is not about phase shifting, or other phase manipulations and sound decomposition.
So maybe I will be more specific (I hope this is not redundant to what I have already written).
---
In general for humans there are several aspects which influence perception of sound direction. Energy of the sound and phase of the sound.

In more details, perception of phase and energy is frequency dependent and perceived in non-linear frequency scale (barks).
- Also perception of energy changes (dynamics) is frequency dependent and nonlinear.
- Perception of phase is related to tonality of certain sound components
(tonality can be represented by something called Unpredictability Measure -> this parameter is used in Aud-X) and frequency dependent as well (for example you cannot determine LFE direction, but you can very precisely of a 2 kHz tone)
--
So the more tonal are the sound components, their phase matters more (but with reference to certain frequency range).
BTW. Tonality descriptor is also relevant in modeling instantaneous masking phenomenon,
but this is a different story.

Anyway, considering above allows to generate a map of energy and phase spectral components and decide which are more significant in terms of perceiving surround panorama, hence which should be considered as more important in generating the 5.1 sound reconstruction scenario (from mono or stereophonic base stream).

The more bitrate is available, the better reconstruction can be performed.
------
I hope (but I am not sure) this answers your last question.

====================

I do not know what is the Kinyo device, how it works and therefore I do not understand why it produces different results than a normal speaker set.

Our informal subjective tests performed on 5.1 speaker set indicated that people were not able to distinguish STDQ Aud-X sound from 448 kbps AC3.
This does not refer to situation where single channel streams are reproduced separately as monophonic, or in pairs - stereophonic. In such a case there is a difference between Aud-X STDQ and AC3 448 kbps.
We have also ran some objective experiments, which we included in the whitepaper.

Ofcourse I will be happy to answer more questions.

Rockaria

25th January 2006, 20:35

I really had a hard time to establish a connection between the crosstalks(fact) and your DPL II.:D
Just regard the kinyo as normal 5.1ch analog speaker. I am also informed the crosstalks from other users. So it's not a matter of specfic speaker model to streighten it. Rather it seems to be a matter of the codec & analog speaker environment overall.

I've heard something similar with the dolby headphone, but because of too deep professional norms considering the written data and my lack of hobby time, I am relying on some future free time and research.

Thanks.

SeeMoreDigital

25th January 2006, 20:43

Is anybody able to provide some samples please?

Cheers

3dsnar

26th January 2006, 07:28

Why do I get confused all the time? ;)

I am not sure if this was what you desired, but:
Are you interested in Aud-X examples?
If yes, here is the instruction how to generate them

1) Obtain an input AC3 file for tests
2) Decode it to PCM with BeSweet
3) Feed the PCM to Aud-X executable encoder
--
If you want to convert Aud-X stram back to PCM:
1) Feed the file to the executable decoder
2) Choose PCM as output format
3) Wait for results
-----
You can also transcode Aud-X stram directly to AC3, if you wish to do so.
=======================================================
All this is very simple, but in case of questions I'd be happy to help.

scharfis_brain

26th January 2006, 11:12

@3dsnar: I have a idea for ultimative compatibility to standard MP3 (or other codecs) decoding:
- while Aud-X encoding downmix the 5.1 signal into two channels using a DPL2 matrix
- then build a 5.1 separation bitstream according to this downmixing.

this will allow users without the Aud-X-Decoder to receive a completely transparent DPL2-Stereo, while Aud-X users will enjoy the advantages of much better channel separation.

3dsnar

26th January 2006, 11:27

Sharfis, thanks for your thought.
Maybe this is a good idea.
I must investigate this issue.
Probably encoding with DPL2 results in stereo stream which is not necessary a linear downmix (with certain weights applied to each channel), but rather a nonlinearly (phase and amplitude) distorted downmix.
---
If my speculation is correct, than marrying Aud-X and DPLII is not possible.
And I am afraid my intuition leads me to the right direction... :(
---
But your post gives me an idea :)
To include real time DPLII transcoder, which would
transcode 5.1 stream (obtained after decoding Aud-X) to
DPLII for those who want to hook the computer up to the external DPLII decoders.
--
We have already implemented AC3 transcoding for external Dolby Digital compatible
home theatre systems, but actually I did not think
of DPLII compatible equipment...

Do you think it is worth considering such an option in our DS filter?

scharfis_brain

26th January 2006, 11:58

DPL and DPL2 are linear downmixes. BUT the sourround channels are encoded with less left/right separation
They have a total phase of 180° between Lt and Rt. The only differnence are their weights.

I do not see the (technical) difference between your downmixing:

Lt= L+C+Ls+LFE
Rt= R+C+Rs+LFE

to DPL2 downmixing:

Lt= L + C + 0.8Ls - 0.5Rs + LFE
Rt= R + C - 0.5Ls + 0.8Rs + LFE
(the weights aren't totally correct here. look at the dolby specs, please)

the result is the mentioned 180° phase difference to restore the surround channels with an additional 180° phase rotation at the decoders side.

maybe it is more easy for you to stay for a DPL1 endoding like this:

S=Ls+Rs
Lt=L+C+S+LFE
Rt=R+C-S+LFE

and then use a 2-step bitstream channel separation:
1st step: separate L, C, R and the mono surround
2nd step: separate Ls and Rs from the mono surround

I hope this could be implemented into Aud-X endoding because it would help it being much more widespread cause it then contains standard DPL(2) which ca be played back at every equipment without Aud-X-decoder without loosing the surround info.