ffdshow tryouts project: Discussion & Development

[tal.aloni]

Hey albain, thanks for your efforts, I'm very happy with your latest build. all mapping problems are gone. :)

I've tested several 5.1 AC3 & E-AC3 tracks, and haven't had a single problem.

a few notes:
1. this may be on purpose, but on the list of Codecs, I only have E-AC3. this may confuse some.

2. I've tried testing some 7.1 E-AC3 tracks from the Dolby Demonstration Blu-Ray, they all play as 5.1. (side surround merges with rear surrounds).
Sonic Cineplayer played them as 5.1 as well.
Power DVD played them as 7.1 just fine in blu-ray mode, and player them as 5.1 in single file mode.
This may have to do with the demuxer and not with the decoder

Thanks for the amazing work!
Tal Aloni

[nautilus7]

2. If you 're talking about E-AC3, libav doesn't output 7.1 yet. Merging the 2 extra channels into the surround ones, is the normal behavior of any E-AC3 decoder not capable of outputting 7.1.

[tal.aloni]

2. If you 're talking about E-AC3, libav doesn't output 7.1 yet. Merging the 2 extra channels into the surround ones, is the normal behavior of any E-AC3 decoder not capable of outputting 7.1.

Again, Thanks for the clarification.

5.1 is just fine for now. :)

[albain]

Has this been reported to ffmpeg dev team?

Not yet, I have troubles to post on their mailing list. If someone can do this, it would be great.

Concerning AAC, I will try to digg this more but the samples that I tested successfully a few days ago on libavformat branch no more work with my last builds, don't know why

[tal.aloni]

Albain,
Unfortunately, now (beta 5) flac (which utilize libavcodec) have wrong channel mapping!

[jruggle]

2. If you 're talking about E-AC3, libav doesn't output 7.1 yet. Merging the 2 extra channels into the surround ones, is the normal behavior of any E-AC3 decoder not capable of outputting 7.1.
A little clarification:
If the 7.1 E-AC3 you have is from Blu-ray, I believe it is usually encoded as a 5.1 pre-downmixed base AC3 stream mixed with a 4-channel supplemental E-AC3 substream which replaces the surround channels and adds the 2 back channels. The lavc decoder skips the supplemental substream, so the result is indeed 5.1 with the extra channels mixed-in, but it's from the original encoder, not from the decoder.

Full 7.1 output will eventually be supported in lavc.

[darbid]

Hello, I have been using ffdshow with WMP(media centre) for over a year now. One of the most annoying things about windows is how it deals with codecs. Direct Show can be misled, currupted or changed so easily.

So the problem is I have no sound when I play DVDs, just the picture. Ffdshow (audio/video) does not pop up in the system tray either.

Using graphedit I have checked which audio/video filter it uses for a VOB file and it has ffdshow (audio and video). I can even play that file with graphedit and ffdshow audio and video pops up in the system tray confirming it is being used.

With a filter manager i have even checked the merits of ffdshow and it is preferred.

Then if i play the same dam VOB file with WMP as soon as the film starts, no sound, no ffdshow audio is started, i just get picture. It must be using another filter for the picture as well.

Could someone please help.

EDIT: Even Gspot will render the video and audio of a VOB (mpeg) with ffdsow. So I imagine I have done something to WMP or Windows itself.

[albain]

Albain,
Unfortunately, now (beta 5) flac (which utilize libavcodec) have wrong channel mapping!

My mistake, I forgot some necessary updates

[Mercury_22]

@albain Can you post a x64 version of beta 5 ? PLEASE ! :thanks: :helpful:

[pitch.fr]

I doubt that it is a bug in FFmpeg. Wasn't it a binutils bug? Just disable whatever function is crashing.
OK thanks for your help clsid, always appreciated :D

yesgrey3 will update ff_samplerate.dll with the latest libsamplerate version then :sly:

[clsid]

yesgrey3 will update ff_samplerate.dll with the latest libsamplerate version then :sly:
Not needed. I have uploaded it here (http://tinyurl.com/3dg). :p

[pitch.fr]

Not needed. I have uploaded it here (http://tinyurl.com/3dg). :p
w00t, thanks!!!!

it's so great to have an awesome tool such as ffdshow maintained by such a knowledgeable and friendly person as you......I envy you actually, you're so awesome :D

[rack04]

OK thanks for your help clsid, always appreciated :D

yesgrey3 will update ff_samplerate.dll with the latest libsamplerate version then :sly:

Sorry if I'm slow but what bug is this referring too?

[clsid]

I don't need a mirror. So I don't have a reason to click on your bogus link.

Rule 14: Multiple registrations are prohibited and are grounds for immediate account deletion.
Can a mod please ban pitch.fr (aka leeperry)?

[clsid]

Sorry if I'm slow but what bug is this referring too?
None. He is just annoyed that I ignored him.

[pitch.fr]

I don't need a mirror. So I don't have a reason to click on your bogus link.


Can a mod please ban pitch.fr (aka leeperry)?
I'm your biggest fan, besides I'm not leepiryy(who's that :confused: )

I wonder why you don't wanna update libsamplerate ?

[pitch.fr]

Sorry if I'm slow but what bug is this referring too?
well, there's been a major update of libsamplerate but the man doesn't care :
http://www.mega-nerd.com/SRC/history.html

what's the point to update ffdshow after all :D

thankfully, yesgrey3 will update the DLL :)

PS: problem solved on my end, have a wonderful day :

clsid has now been successfully added to your ignore list.

[Kurtnoise]

DVD subs (aka .idx/.sub) are bitmaps. If you want to change the font, you'll have to OCR all subtitle images, possibly re-position them and save them in some other format or create new subtitle images.

I think you'll agree that this is a wee bit more complicated than "just changing a setting"...


Sounds like you haven't tested SC yet...no, it doesn't use OCR engine for this part.


to stay on the topic, I can confirm that the AAC lavc decoder doesn't work with AAC streams embedded in mp4 container using the last beta from albain.

[yesgrey]

OK thanks for your help clsid, always appreciated :D

yesgrey3 will update ff_samplerate.dll with the latest libsamplerate version then :sly:

pitch.fr,
Please put me out of your fights, I will not allow you, or anyone, to use me to provoque anyone else. I have PMd you saying that I will try to update the libsamplerate because I did not agree and did not like clsid's position, but now, publicly, I have to tell you that I did not like that you have come here and have used me to attack clsid. I am not here to fight or to help anyone in a fight. I am here willing to learn, to give some suggestions, and trying to help whenever I can, without expecting to receive anything but respect.
I will still try to update and post here the new code for ffdshow's ff_samplerate.dll; not because of you, but because I feel that everyone who wishes to try and use the new version, besides the increased size, should be allowed to do it. If the developers of ffdshow tryouts decide to accept the updated code and use it it's their decision, because it's their software.

Not needed. I have uploaded it here (http://tinyurl.com/3dg). :p

clsid,
I have learned to respect you during the time I have spend here, but this "words" of yours reduced it considerably.

Sorry for the off-topic, but my name has been introduced without my consent.

[pitch.fr]

pitch.fr,
Please put me out of your fights

humm ?
you told me yesterday in PM that you were awaiting Mr Big Stuff's answer on whether he was gonna update it.

and because he prefered to ignore OUR(you, richardk and I) request from the previous page, you'd try to do it.

anyway, I wanted to let him know that he can ignore OUR request as much as he wants.....it'd still be done, thanks to your mad coding skills :rolleyes:

thanks for your endless efforts yesgrey, but rejoice.....we got a new Reclock developer, how cool is that :eek: :D

I'm not the one who used the F word here, if you look closely ;)

I'm a reasonable person, not the case of everyone around here....it seems :rolleyes:

[albain]

Wow, there was some interesting debates...

Getting back to serious things : AAC decoder does not work or crash when compiling FFDShow (not libavcodec) under release mode. In debug mode it works. When it crashes it is raised from libavcodec dll. Weird...

Let's hope that the upcoming major updates of AAC decoder will solve this problem.

I will post an updated version (32 & 64 bits) very soon

[STaRGaZeR]

My experience with beta5:

1 - E-AC3 decoder works fine :)
2 - Using livabcodec for normal AC3 decoding sometimes makes ugly high pitch sounds that last miliseconds with AC3 in AVI and when playing DVDs with MPC-HC. Matroska is fine so far, but it'll be nice to separate E-AC3 from normal AC3 so we can select libavcodec and liba52 respectively.
3 - AAC decoder does not work, like the others have said.
4 - libavcodec DTS crashes always. libdts is perfectly fine though.

TrueHD would be a nice addition now.

[rack04]

4 - libavcodec DTS crashes always. libdts is perfectly fine though.

I can't get any codec to work with DTS. Well, correction they work, but the audio is screwed. Sorry don't know of a better way to explain it. Is libav recommended for all stable audio formats?

[yesgrey]

I have updated here (http://forum.doom9.org/showthread.php?p=1186274#post1186274) ffdshow's libsamplerate to version 0.1.4.

If you want, try it and let us know what you think about it.

[foxyshadis]

I'm your biggest fan, besides I'm not leepiryy(who's that :confused: )

I wonder why you don't wanna update libsamplerate ?

I think you are. I've been looking through both accounts, and I'm ready to ban you for skirting your 30-day suspension. You're provocative enough in this thread to earn a rule 4 anyway.

yesgrey, clsid, I believe it should be offered as an optional package, along with other things that are rarely used, updated, or outdated: libmad, realaac,& tremor (all need to be removed generally anyway, libavcodec is far superior), plus theora, and xvidcore.

[Lux Delux]

Just an observation, as suggested by Lord Mulder ( ;) ) I found this thread, with ffdshow_rev2135_20080921_clsid.exe I got weird blocking in a brighter area

I'm using megui with the latest skystrife build, just now I noticed (in just 1 scene for now) a strange blocking in an area

http://img217.imageshack.us/img217/1049/snapshot20080922004609uq2.jpg (http://imageshack.us)

below is the 1st and 2nd pass settings


Job commandline: "C:\Program Files (x86)\megui\tools\x264\x264.exe" --pass 1 --bitrate 1137 --stats "D:\WALKER S1\01\udegrain2.stats"
--ref 16 --mixed-refs --no-fast-pskip --bframes 6
--b-adapt 2 --b-pyramid --b-rdo --bime --weightb --direct auto --filter -1:-1 --subme 7 --trellis 2
--partitions p8x8,b8x8,i4x4,i8x8 --8x8dct
--me tesa --merange 32 --threads auto --thread-input --sar 59:65 --progress
--no-dct-decimate --no-psnr --no-ssim --output NUL "D:\WALKER S1\01\udegrain2.avs"

Job commandline: "C:\Program Files (x86)\megui\tools\x264\x264.exe" --pass 2 --bitrate 1137
--stats "D:\WALKER S1\01\udegrain2.stats"
--ref 16 --mixed-refs --no-fast-pskip --bframes 6
--b-adapt 2 --b-pyramid --b-rdo --bime --weightb --direct auto --filter -1:-1 --subme 7 --trellis 2
--partitions p8x8,b8x8,i4x4,i8x8 --8x8dct
--me tesa --merange 32 --threads auto --thread-input --sar 59:65 --progress --no-dct-decimate --no-psnr
--no-ssim --output "D:\WALKER S1\01\udegrain2.mkv" "D:\WALKER S1\01\udegrain2.avs"


with the official beta release everything is fine. BTW keep up the good work :)

[Liisachan]

other things that are rarely used, updated, or outdated: libmad, realaac,& tremor (all need to be removed generally anyway, libavcodec is far superior), plus theora, and xvidcore.

Don't remove xvid as vfw decoder please. libavcodec is NOT superior, it has b-frame delay, making frame number messed up when ffdshow is used by software like VirtualDub, av sync problems are possible too.
Also last time I tested Tremor was better than lavc for multichannel Vorbis. Multichannel Vorbis is only rarely used and honestly I'm not interested in this bug, but the point is, don't believe in lavc blindly. It's not like libavcodec is superior in everything. I agree with you about libmad however.

[albain]

Here is the updated versions of the branch for AC3/EAC3 decoding :

FFDShow 32 bits
http://rapidshare.com/files/147368880/ffdshow_rev2134_20080922_dbt_beta7.exe.html

FFDShow 64 bits
<On the way>

Nothing new on DTS and AAC decoding

[Mercury_22]

Here is the updated versions of the branch for AC3/EAC3 decoding :

FFDShow 32 bits
http://rapidshare.com/files/147368880/ffdshow_rev2134_20080922_dbt_beta7.exe.html

FFDShow 64 bits
<On the way>

Nothing new on DTS and AAC decoding

Thanks ! Waiting for the 64 bits :) :thanks: :helpful:

[clsid]

yesgrey, clsid, I believe it should be offered as an optional package, along with other things that are rarely used, updated, or outdated: libmad, realaac,& tremor (all need to be removed generally anyway, libavcodec is far superior), plus theora, and xvidcore.The only reason I was against updating libsamplerate is because of its radical increase in size. The large size itself isn't the biggest issue. It is the 900% increase that makes me suspect something is wrong. Maybe GCC is able to do a better job. But atm I can't be bothered to try it, maybe someone else here wants to.

Realaac can indeed be removed as I am not aware of any samples that fail with libfaad2 and work with realaac.

I don't like the idea of multiple versions or an addon pack much. Most of the non-essential components aren't very large in size comapred to the core components.

Nothing new on DTS and AAC decodingDoes the DTS decoder always crash in the same function? If so, just disable that function in libavcodec.

[fastplayer]

Why remove codecs from ffdshow at all? Is it that difficult to maintain them? I mean it's not like they're updated on a daily basis (unlike ffmpeg code). The Xvid guys maybe release an update once a year...
Having the option to choose from multiple codecs for some of the A/V formats is definitely an advantage. Especially at zero cost...

[clsid]

The cost is bandwidth/diskspace. There is only an advantage is one decoder can do things that its alternative can't.

[yesgrey]

The only reason I was against updating libsamplerate is because of its radical increase in size. The large size itself isn't the biggest issue. It is the 900% increase that makes me suspect something is wrong.

From inspecting the code, the large size increase is due to the new files high_qual_coeffs.h and mid_qual_coeffs.h; the size went from 486kB and 151kB, respectivelly, to 8973kB and 593kB!!! I think there is nothing to do to decrease the size, these coefficients are all hardcoded in the dll...

I also think there is nothing wrong, it was the way the author find to increase the quality of the best two modes. See the differences below (these numbers are described in the source code)

Old Resampler best three modes:
1: 97dB SNR, 96% BW
2: 97dB SNR, 90% BW
3: 97dB SNR, 80% BW

New Resampler best three modes:
1: 145dB SNR, 96% BW
2: 121dB SNR, 90% BW
3: 97dB SNR, 80% BW

After all this, what is really important is the answer to the question:
Is the new best mode of libsamplerate better than the best mode of libavcodec?
If it's not, it's not worth it to update the code. In fact, if it's not, we should also consider removing libsamplerate, because there is no need to have something that is of lower quality...

I will try to test both carefully and will post here the results. It would be great if more people test the sound quality and post here the results.

[madshi]

After all this, what is really important is the answer to the question:
Is the new best mode of libsamplerate better than the best mode of libavcodec?
If it's not, it's not worth it to update the code. In fact, if it's not, we should also consider removing libsamplerate, because there is no need to have something that is of lower quality...

I will try to test both carefully and will post here the results. It would be great if more people test the sound quality and post here the results.
Would it also be possible for you to do a cross comparison to the free r8brain? I've always wondered how good r8brain is compared to libsamplerate...

http://www.voxengo.com/product/r8brain/

[albain]

Does the DTS decoder always crash in the same function? If so, just disable that function in libavcodec.

I am unable to detect in which function it crashes as it does crash only in release mode, not in debug mode.
I will try to play with the compilation flags in order to get debugging + crash

[clsid]

I suspect one of the float_to_int16_interleave_* functions from dsputil_mmx.c crashes. Try disabling those.

[yesgrey]

Would it also be possible for you to do a cross comparison to the free r8brain? I've always wondered how good r8brain is compared to libsamplerate...

http://www.voxengo.com/product/r8brain/

It's funny, A few days back I've suggested pitch.fr to do exactly that, because I thought that if you have decided to use r8brain in eac3to was because you thougt it was the best... now I know we are in the same boat: we don't know. :)

I will compare all three.

[madshi]

It's funny, A few days back I've suggested pitch.fr to do exactly that, because I thought that if you have decided to use r8brain in eac3to was because you thougt it was the best... now I know we are in the same boat: we don't know. :)
You're quite right - I don't know!! :o I had read an article at some time comparing the professional (non free) r8brain version to libsamplerate and several other free and commercial resampling applications, and r8brain came out very good. But the free version is not as good as the commercial one. So I never knew for sure.

I will compare all three.
Great - thanks much! Looking forward to your results...

[albain]

The DTS decoder crashes in function ff_imdct_half_sse.

It works when disabling it

[tal.aloni]

Albain,
In beta7, I'm getting the correct channel mapping for all of the formats! (Flac, AC3, E-AC3)

We finally have 100% Free and 100% Working Direct Show E-AC3 Decoder! That's amazing!

Thanks a lot!
Tal Aloni

[Inventive Software]

Anybody got any samples and a recent build I can test this with? I'm eager to try this out. :)

What's the status on MLP / TrueHD decoding?

[ACrowley]

Albain,
In beta7, I'm getting the correct channel mapping for all of the formats! (Flac, AC3, E-AC3)

We finally have 100% Free and 100% Working Direct Show E-AC3 Decoder! That's amazing!

Thanks a lot!
Tal Aloni

yeah...fine

But what about 24bit output ? i get alwys 16bit ? Shouldnt it be 24bit ?
eac3to output with NeroAudioDecoderII and Sonic decoder is 24bit

[albain]

yeah...fine

But what about 24bit output ? i get alwys 16bit ? Shouldnt it be 24bit ?
eac3to output with NeroAudioDecoderII and Sonic decoder is 24bit

Is it a problem with the decoder or with FFDShow : I mean should the output format be configured into the decoder itself or else when the output format selected in FFDShow is 24 bit, it would not be taken into account by the decoder ?

Concerning 64 bit build, I have troubles to compile it since I switched to the branch (problem with yasm on imgfilters/skl..., I will spare more details). As soon as it compiles I will post updated versions with the working DTS decoder too

[Mercury_22]

Is it a problem with the decoder or with FFDShow : I mean should the output format be configured into the decoder itself or else when the output format selected in FFDShow is 24 bit, it would not be taken into account by the decoder ?

Concerning 64 bit build, I have troubles to compile it since I switched to the branch (problem with yasm on imgfilters/skl..., I will spare more details). As soon as it compiles I will post updated versions with the working DTS decoder too

Why not switching back ? I think it's time / good (enough) to go back to main ! :thanks::helpful:

P.S. We can keep liba too (just in case) but why not making it official ?:confused:

[shambles]

Is it a problem with the decoder or with FFDShow : I mean should the output format be configured into the decoder itself or else when the output format selected in FFDShow is 24 bit, it would not be taken into account by the decoder ?

i think it's a problem with the decoder itself.. selecting 24bit output in ffdshow and dumping output creates 24bit files but with the last 8 bytes padded.

madshi has hacked the libavcodec (e-)ac3 decoder for eac3to (http://forum.doom9.org/showthread.php?p=1094167#post1094167) so it outputs floating point and that is then dithered down to 24bit

[albain]

This is not my call, and there is still work to do concerning TrueHD decoding, I didn't have the time to work on it yet

[albain]

Here are the updated versions with EAC3/DTS decoding :

32 bits :
http://rapidshare.com/files/147650229/ffdshow_rev2138_20080923_dbt_beta8.exe.html

64 bits :
http://rapidshare.com/files/147650547/ffdshow_rev2138_20080923_x64_beta8.exe.html

[Mercury_22]

Here are the updated versions with EAC3/DTS decoding :

32 bits :
http://rapidshare.com/files/147650229/ffdshow_rev2138_20080923_dbt_beta8.exe.html

64 bits :
http://rapidshare.com/files/147650547/ffdshow_rev2138_20080923_x64_beta8.exe.html

64 bits it's crashing with Libav for E-AC3 / AC3 / DTS tested with WMP 64-bit and MPC-HC 64-bit
Samples: E-AC3 (http://rapidshare.com/files/143609863/Test_E-AC3.ts.html), AC3 (http://rapidshare.com/files/143875772/Test_LibavAC3.avi.html), DTS (http://rapidshare.com/files/147667691/Test_DTS.ts.html)
Problem signature:
Problem Event Name: APPCRASH
Application Name: wmplayer.exe
Application Version: 11.0.6001.7000
Application Timestamp: 47919dbe
Fault Module Name: libavcodec.dll
Fault Module Version: 0.0.0.0
Fault Module Timestamp: 48d8b260
Exception Code: c0000005
Exception Offset: 000000000034bec8
OS Version: 6.0.6001.2.1.0.256.1
Locale ID: 1033
Additional Information 1: 38fa
Additional Information 2: ac7e3f1ebb373830e789317be6c118ae
Additional Information 3: f834
Additional Information 4: 455055a6a1260056002639897fbd0fa0
:helpful:

[clsid]

Why not switching back ? I think it's time / good (enough) to go back to main !#$%@^$^%$@$@$% *deep breath* *deep breath* !$#%@@!@

Several problems have been reported that have not yet been fixed afaik. AAC is still completely broken. Does 2 channel AC3 work yet? What about the AC3 decoding quality issues that were reported?

Are AC3 and E-AC3 configurable separately yet? NO.
Have things been properly tested yet? NO.

[fastplayer]

clsid, please watch your blood pressure! :thanks:
:D