DBZ audio sync issue, remastered sets. different delays NTSC DVD

[xekon]

(The vob I gave you I just copy and pasted from the DVD with zero processing...)

My original method that I have been working with up until yesterday was with Honor Pulldown. That is why you can see interlacing on my picture of the eycatch on page 2. I did notice the amount of delayaudio I had to use was different with Honor Pulldown than it was with Force Film(-0.165 vs -0.141), and I was wondering why that was, thanks for that explanation. The reported delay to use for the episode is -0.213 but that throws the entire episode out of sync

I must have misunderstood, I thought when you asked for the steps that I took that you wanted to all of the steps I took from step A to step Z.
(In the previous steps I outlined, for my process I was letting PGCDemux strip the angle.)

Please let me know if the following would be ok:
(and I'll do it over, the problem is that I do not know exactly what you need because you did say unprocessed.)

My steps:
1. pull 1st .VOB directly from DVD (without decrypting, not using anydvd, dvddecrypter, or ripit4me)

2. run that .VOB through IfoEdit to remove all but the first angle.

3. Upload that .VOB to you.

Then once you download the files the only steps to follow from my original guide:

6. Load DGIndex, set Video > Field Operation > "Honor Pulldown Flags"
Open VTS_12_1.VOB and goto File -> Save Project, which will generate a .d2v

7. Load AvsPmod, use the following test.avs:


LoadPlugin("F:\AviSynth\plugins\audgraph.dll")
LoadPlugin("F:\AviSynth\plugins\DGDecode.dll")
LoadPlugin("F:\AviSynth\plugins\NicAudio.dll")
A = NicAC3Source("F:\s9d1\001.ac3")
V = MPEG2Source("F:\s9d1\001.d2v")
AudioDub(V, A)
DelayAudio(-0.165)
ConvertToRGB()
AudioGraph(3)

[Guest]

Perfect, thanks. I can work on it this evening.

Can you please give me the frame numbers for the events you point to again but for honor pulldown? They will be different from the numbers for forced film.

[xekon]

yes thats why I deleted those comments from the original guide. I needed to double check the frame #'s

I will get to work on it right now :) thanks again. Also did mediafire DL at a good rate or would you prefer rapidshare?

[Guest]

Mediafire worked great for me. Don't ever point me to rapidshare!

I can't work on it until this evening so you have a good 6 hours to collect new frame numbers for the critical sync points.

[xekon]

If I use DGIndex on the .VOB directly the amount of delay to get it back into sync will be different than the amount of delay to get it back into sync if I use PGCDemux to demux by Vob ID.

This is because there are audioless cells before every episode.

would you like the me to use PGCDemux to demux by VOB ID and give you that delay + Frame #s for sound events.

or would you like me to use DGIndex on the .VOB directly and adjust the delay and Frame #s to line up?

Also if you prefer I can remove the audioless cells but then the amount need to line everything up is a positive delay, roughly +0.048 for the first episode, and then by the time you get to the 6th episode on the disc its around 0.334

or another option since you will only be working with the first episode I could use the right arrow a couple times then use the '[' button to cut off several black frames at the very beginning in DGIndex and report those Frame #s plus delay.

[Guest]

or another option since you will only be working with the first episode I could use the right arrow a couple times then use the '[' button to cut off several black frames at the very beginning in DGIndex and report those Frame #s plus delay. Yes, do that. Please report the number of > hits before the [. Then report the frame numbers for the key events as seen in AvsPmod or VirtualDub.

[xekon]

neuron2, I am still working on uploading the sample files, I will send a PM once they have finished uploading.

1. Load DGIndex, set Video > Field Operation > "Honor Pulldown Flags"
Open VTS_12_1.VOB

click the "[" key here:

http://i1208.photobucket.com/albums/cc361/xekon/dga_zpsb4241048.png

2. Load AvsPmod, use the following test.avs:


LoadPlugin("F:\AviSynth\plugins\audgraph.dll")
LoadPlugin("F:\AviSynth\plugins\DGDecode.dll")
LoadPlugin("F:\AviSynth\plugins\NicAudio.dll")
A = NicAC3Source("F:\samp\VTS_12_1 T81 2_0ch 192Kbps DELAY -213ms.ac3")
V = MPEG2Source("F:\samp\VTS_12_1.d2v")
AudioDub(V, A)
DelayAudio(-0.165)
ConvertToRGB()
AudioGraph(6)


goto Frame 16747/16748 and you will see the sharp change in the audiograph with the scene.
also at Frame 1928 he clicks the trigger, you can see the difference in the audiograph
also at Frame 0 with DelayAudio(-0.165) the first frame is lined up with the audio,
if you comment out DelayAudio(-0.165) you will see that the audio is off by roughly 5 frames.
and that a delay of -213ms would lose some of the audio that belongs with the first couple frames of the episode.

Now something else I noticed, if you click ">" only 5 times from the very start of the .VOB and save project, the .ac3 file that DGIndex demuxes reports a delay of -165ms. I found this interesting because thats the delay that works with the episode. but if you try and use this .d2v with the .ac3, the audio is not actually a delay of -165ms. but I did find that -165ms value interesting... like maybe its not a coincidence, but maybe it is.

I tried doing this with the second episode (click ">" only 5 times from the very start of the .VOB), and DGIndex reports a delay of -165ms for the second episode too, but the second episode has a different delay than the first episode, so its either just a coincidence or I just cant figure out the significance.

[Guest]

Whoa! What sample files? I already downloaded one set.

Using the VOB I have already downloaded, referring to the timestamps dump I generated with DGIndex, one sees that you need to skip 4 GOPs at the start. So, hit > 4 times and then [. Then save project. When I do that I get AC3 files with 0 delay reported. I then make this script:

vid=mpeg2source("VTS_12_1.d2v")
aud=nicac3source("VTS_12_1 T82 1_0ch 96Kbps DELAY 0ms.ac3")
audiodub(vid,aud)

Sync looks fine.

Please don't make me do big downloads repeatedly. Let's work with this one I have if possible.

It appears that the only problem here is the starting GOPs without audio.

[xekon]

Oh ok, I was going to reupload so I could strip the angle for you. Can you see if sync Looks good on T81 instead of T82?

I am going to checkout T82 now to see if I see the same difference that you do.

[Guest]

You have multiangles in the VOB you gave me? I'll wait for your clarification before proceeding. I will have to get the new sample if the original had multiple angles.

Anyway, it's all about correctly skipping the leading GOPs without audio. Use the timestamps dump to determine how many to skip. If you skip too many, then you get to an open GOP and will have a nonzero delay.

[xekon]

Yes, I gave you a multiangle VOB because i'm still a noob, and when you said you wanted the VOB before any processing I took that literally and did nothing with the VOB before sending it to you.

(The new one that I was working on uploading has only had one single processing done, and that was using IfoEdit to strip the angle.)

I also get a 0ms reported delay when I use the ">" four times before saving.

but the audio delay that it takes to fix it if I do that is DelayAudio(1.283)

http://i1208.photobucket.com/albums/cc361/xekon/fixa_zps00413717.png

next frame:
http://i1208.photobucket.com/albums/cc361/xekon/fixb_zpsd3732bfa.png

[xekon]

vid=mpeg2source("VTS_12_1.d2v")
aud=nicac3source("VTS_12_1 T82 1_0ch 96Kbps DELAY 0ms.ac3")
audiodub(vid,aud)


Edit: It looked like Track 82 was lining up great, because the audio graph was in sync, but it is actually waaaaaay out of sync, and was lining up with something else, once I played it back and listed to it, it was very obvious. For a minute there it looked like Track 82 was syncing up better than track 81.

(The track I am wanting to use is Track 81.)

[Guest]

Can you please answer my question? Is the VOB you gave me multiangle? I have to know before I can do anything else.

[xekon]

very sorry, Yes it is multiangle. The part that is multiangle is the first part of the episode where it says the name of the episode.

[Guest]

OK, no problem. Can you please provide a stream ripped with a single angle? My idea is to have a common starting point eliminating any unnecessary variables.

[xekon]

Yes I have it uploading now. It is 1GB like the last, and my connection is only Docsis 2.0 so it will probably be about an hour before I have it all finished uploading.

I really wish they would upgrade the cable service in my area, the next town over where my aunt lives has Docsis 3.0 and it is sooooooo much faster.

[Guest]

You know what they say, patience is a virtue. And it gives me some time to sip a little beverage while contemplating the idiocy of modern physics.

This is a cool thread because it is exposing some of the intricacies of AV sync in anime. Even how to define AV sync in anime raises interesting philosophical questions.

[xekon]

Okay, the one hour was a guess on my part by how long it took to upload the first .part file. I then stopped because I wasnt sure if you needed me to upload the rest.

This time I loaded all remaining 5 files for upload and it is telling me a total of 5 hours, so if you have anything else to be doing please do not wait up for me.

hate my crappy internet lol.

[xekon]

You have PM for the new Links, files just finished uploading.

[Guest]

OK, I have loaded the VOB, hit > four times, then [ and save project, with honor pulldown flags. I have this script:

v=mpeg2source("VTS_12_1.d2v")
a=nicac3source("VTS_12_1 T80 3_2ch 448Kbps DELAY 0ms.ac3",2)
audiodub(v,a)

Sync looks great. Please advise.

[xekon]

ok let me try With Track80. (although the track I am wanting to use is Track 81.)

If Track 80 is in sync and Track 81 is not, what does that mean?

[Guest]

I checked T81 and it was fine too.

Please point me to any issues you see by frame number.

[xekon]

ok with track 81, if you playback this section:

Frame 18363 to 18578

where the old man is talking to Gohan, a couple words worth of voices come out of his mouth before his lips even begin to move, and then after the old man is done talking, his lips are still flapping for about a couple words worth.

then gohan goes "uhhhhnnn" like how did he know what I was thinking. This is supposed to happen with gohans face on the screen, but it actually happens before it has switched from showing the old man to gohan.

but if I use DelayAudio(1.283) then everything is in alignment like it should be.

(If all of this lining up for you and not me, I cant think of anything we could be using differently, I am using the same sample files that I uploaded)

I am using Media Player Classic Home Cinema to play back it back, and I have combined community codec pack installed on default settings.

LoadPlugin("F:\AviSynth\plugins\audgraph.dll")
LoadPlugin("F:\AviSynth\plugins\DGDecode.dll")
LoadPlugin("F:\AviSynth\plugins\NicAudio.dll")
A = NicAC3Source("F:\samp\VTS_12_1 T81 2_0ch 192Kbps DELAY 0ms.ac3")
V = MPEG2Source("F:\samp\VTS_12_1.d2v")
AudioDub(V, A)
DelayAudio(1.283)
ConvertToRGB()
AudioGraph(6)

[Guest]

Sorry I am not going to get involved in arguing about lip sync for dubbed audio speech. I looked at some other clear AV sync events and they lined up fine. If you like I can point them out.

Here's one in the section you mentioned, look at 18378 transitioning to 18379. Just as the mouth moves the audio arrives in the center of the audio graph.

If there are other sections where this is not the case, it just underlines what I said earlier about some sections being "in sync" and others not. So which is the "correct" offset?

Edit: We must have some basic difference because the scene shifts away from the man at 18508, but you talk about it up until frame 18578. You should NOT apply the DelayAudio(). Why are you doing that??? You really think that the audio is off by 1.28 seconds???

Are you using the same process and script as I have? Please use VirtualDub to open the script and view frame numbers. Obviously if you use DirectShow codecs for playback you are going to get incorrect results.

[xekon]

The audio is sooooooo far off that the problem is obvious. If I pop in the original DVD into my DVD player then this problem does not happen.

were you able to get the audiograph to line up in the center?

I completely understand if you tell me that this issue is not happening for you at all, and that you are done trying to help. In fact if everything is perfect for you then I probably need to think about switching media players or computers and try and figure out what the variable is between our two setups.

please do not just conclude with not going to get involved, I just need a simple answer of weather or not you are seeing the same problem that I am, then atleast I will have a little confirmation of what I need to look into next. a delay of 1.283 is Wild when the delay should have been 0ms, were talking about a huge difference in sync here.

http://i1208.photobucket.com/albums/cc361/xekon/fixa_zps00413717.png

next frame:
http://i1208.photobucket.com/albums/cc361/xekon/fixb_zpsd3732bfa.png

[xekon]

Edit: We must have some basic difference because the scene shifts away from the man at 18508, but you talk about it up until frame 18578. You should NOT apply the DelayAudio(). Why are you doing that??? You really think that the audio is off by 1.28 seconds???

yes, and my frame numbers are different, i think maybe you noticed something, one sec posting pics

edit: here are three sequential frames: this is the transition from the old man to gohan that i was referring to, does this happen at the same frame numbers for you?

one:
http://i1208.photobucket.com/albums/cc361/xekon/one_zps0d0c87c9.png

two:
http://i1208.photobucket.com/albums/cc361/xekon/two_zpsa6f21c17.png

three:
http://i1208.photobucket.com/albums/cc361/xekon/three_zps900f5958.png

EDIT: trying virtualdub now, I honestly have not used it before may take me a second to figure it out if there are any large differences.

[Guest]

You still have the DelayAudio() call in there. Get rid of it!

EDIT: The scene shifts away at 18508 for me but not for you. Are you sure you have Honor Pulldown? This is our problem and we need to get to the bottom of that first. Please post your D2V file so I can compare them and find out why we have this difference.

[xekon]

you asked if I really think 1.28 seconds is the delay, and yes I do. I have been testing with DelayAudio() commented out and with it there.

When I use the 1.28 second delay the entire episode lines up great.

At this very moment I am working on opening it in VirtualDub, and I am going to have delayaudio() removed, and take new screenshots.

I am very curious to see if my screenshot looks different from your screenshot under the same circumstances.

Please let me know if I missed any questions, I read it over twice, and I am sorry for my unorganized responses at times. I will read things over more thoroughly before posting to make sure I answer any and all points before replying.

[Guest]

Please post the D2V file, as I asked. Thank you.

We have to have our video frame numbers lining up before even thinking about audio.

[xekon]

ah I missed that, thank you, uploading now. Also when DGIndex says it noticed a field transition problem. would you like me to have clicked yes or no?

right now I am uploading both incase it matters.

edit: just sent PM with DL link. I also zipped it with the Full path included so you can see how I had them arranged on my hard drive.

F:\samp\VTS_12_1.d2v
F:\samp\VTS_12_1 T81 2_0ch 192Kbps DELAY 0ms.ac3
F:\samp\VTS_12_1.VOB

[Guest]

I do not get the field order prompt. I looked at the D2V you mailed me and it appears you have not skipped the first 4 GOPs as I described.

Again:

Open the VOB. Hit > four times. Hit [. Then save project with Honor Pulldown.

Is that exactly what you did?

[xekon]

Yes originally that is what I did, however I just recreated these files because I was unsure if I clicked yes to correct field order prompt or not.

One second I am doing it again, very sorry, my brain just has not had the affects of the coffee yet or something.

OK I just deleted my old one, new one is in your PM, and it is the exact same one I will use. (you are correct that with four presses it does not ask to fix the field order prompt.)

New link PM to you.

edit: I got virtualdub and I am using the new .d2v that I just generated and sent to you (deleted my old one.) about to check some frames and then post screenshots.

[Guest]

Ok, thanks. I suggest that you really study the timestamp dump and understand how it tells you the right number of GOPs to skip. I'll be happy to explain it if it is not obvious.

I'll look at the new D2V but I don't expect any surprises.

[Guest]

Our D2Vs agree now so I expect you to confirm my results any second now. :)

BTW, the field order transition is a result of not cutting off the first bad GOPs.

[xekon]

So here is what happened, I have been using sooo many tools and soo many different methods and orders to the point of getting flustered and missing steps that I thought I had done myself.

What I had done was click the ">" four times and then save project without setting the stop. "[" on my last attempt. which caused the delay to be 1.23

but I had audio files that had the " DELAY 0ms.ac3" in the filename, which means I got really mixed up at some point and did it right the first time, but then did it wrong the second time, and then worked with those files which gave me the delay of 1.23

You Sir are correct, the first episode is lining up correctly, the center eyecatch does not line up as perfectly as I thought it would, but that may be the way Funimation did it.

I will do this with the remaining 5 episodes on the disc to see if they all line up this good or if it gets worse per episode.

I believe they should all be good like this episode though! I really hope so that would be great.

[Guest]

It's an important lesson to learn. When debugging or analyzing, always strive to reduce things to the smallest number of variables possible. Easy to say, sometimes hard to do.

I'll look at the eyecatch again on T81. I seem to recall it was within a frame of being correct on T80. Who knows if they timed it right on the original. That's why anime sync is so challenging and your thread is so important, despite some bumps along the way.

[xekon]

Yes very much an important lesson, one that I seem to learn over and over :( I always tend to get ahead of myself or mix things up the point of chaos.

here is the eyecatch for track 81 (I marked with purple how I expected it to line up.):
I expected the part I circled to line up with the Vertical Line that I drew. (but like I said this is a much smaller amount of delay than 1.23 maybe funimation did this on the original mux. I am going to try other episodes and take screenshots of their eyecatch too, that way I can look for differences in alignment.)

http://i1208.photobucket.com/albums/cc361/xekon/align_zps7e65af26.png

[Guest]

Here's my eyecatch on T81. The audio starts exactly at the scene change.

You see it's like I said, some parts in sync, some not. It's apparently the nature of anime dubbing. Chopping the bad GOPs gets you the "real" delay encoded in the stream, but it may not be really real. Now I'm confusing myself :)

...

[xekon]

That is how mine looks at that particular frame as well. but in that scene they may have meant for there to be a 2 frame delay before the audio started for the eyecatch,

but in the screenshot I posted you have some audio that happens before the scene change happens, (where it goes from showing one car to showing a different car)

This to me seemed to show an issue more, however maybe Funimation made this error when they dubbed it.

Earlier in my tests when I was manually lining up episodes the amount of delay that was need to correct the episode varied per episode. The first episode had the least amount of problem, the delay needed was only 48ms while later episodes were off by as much as 170ms.

I am working on checking the other episodes now.

[Guest]

My point is who knows what they intended, or if they erred in implementing their intentions?

Going down to this single frame sync analysis for anime I think is guaranteed to drive you bonkers.

Play it at normal speed and if everything looks fine why worry?

You could study every sync event and try to minimize the number of imputed misalignments, but is it really worth it when played at normal speed, nothing really appears wrong? This is the philosophical issue I alluded too earlier.

In working with the other episodes, be sure to chop the correct number of GOPs. I assume you know how to get that number from the timestamps dump, as you have not asked me to explain that. Just look for how many leading GOPs you have without audio.

[xekon]

I do think I understand now how to chop the correct number of GOPs. Thank you :)

[Guest]

You're welcome. I have to thank you too for bringing that audiograph() function to my notice. I wasn't aware of it and how useful it is.

[xekon]

I have used Blender & Sony Vegas and other non linear video editors before for other projects, and really like being able to see the waveform.

When I started having sync issue with this project I immediately started looking for a way to see the waveform with avisynth :)

[xekon]

Alright so that episode was about 2 frames off. Another one I am doing now is 5 frames off. I am guessing since you could not find a problem with the stream that it is indeed something that was off when it was originally muxed by the producer.

I can see the difference, and with only 1-2 frames off its almost impossible to tell, but with 5 frames off it gets even more noticeable.

Since I cannot figure out what is happening exactly my best idea is to find the alignment of each episode by looking at the center of the eyecatch on every episode and make it so the eyecatch lines up the same for all of them. Then I should be within 3ms for the sync, which should be impossible to notice while watching it. a lot better than 5 frames off anyway :)

I was just really hoping for a way to get the delay information without adjusting manually per episode, but if the problem happened directly from the producer than it is very possible that there is no way of doing that. I have been reading through some forums I found by searching google for a similar issue and found somebody talking about the sync being off on the remastered blu ray version of a DBZ movie but the audio sync was perfect in the DVD version, so that alone tells me that they do make mistakes when they produce these (can I have their job, talk about quality control problem)

[Guest]

Sounds like you've got the bases well covered. I haven't seen your new stream so I'll not comment further.

[xekon]

It seems there production method must have changed as they produced these, the small cell toward the opening of the episode that gives the Name of the episode was film for season 9 but on the older series it is not reported as FILM for that cell. Also the eyecatch is a different scene, it shows little gohan getting on gokus shoulders, and the music is different, it does not have that nice sharp cut off, even though there is sound that I could align per episode to look the same, once I had one to go off of.

I will post more pictures after doing 20+ episodes. Then I will know which episodes show it the most and in the most obvious ways to visually and audibly confirm. (It will probably be a week or so, I want to do a bunch of them the way I described as it will probably end up being the only way to do it.)

[Guest]

You haven't actually demonstrated any audio sync problem yet for a properly handled stream, so you're really talking to yourself now. I will still suggest to just use the encoded AV sync and thereby save a lot of time and agonizing.

[xekon]

Well, that is what I was trying to show with this picture:

http://i1208.photobucket.com/albums/cc361/xekon/align_zps7e65af26.png

Music happens on frames where music does not belong, a delay of 0.048 corrects this. But because the delay is only 48ms the episode look OK to most people because it is such a small difference weather you fix it or not.

I know the difference is small, I actually got about half way through the 291 episodes before I started noticing that some of the episodes did not feel like they were in sync when watching them.

So I started investigating. As I correct them all manually I will store the delay required to fix each episode and list the worse ones after I am done.

[Corpsecreate]

whoaaa so much to read, how did I not see this thread ages ago. I'll pm u xekon with my scripts once I turn my PC on. Also, the episode that will give u the most trouble is episode 65 right before the eyecatch. You'll see what I mean.

[xekon]

hey, welcome to the thread, I will have to check that out.

some discussion about the sync issue with the dvds that I am working with, the problem is in the source, and there is nothing i can do about it.
http://www.kanzenshuu.com/forum/viewtopic.php?f=7&t=16542

The original boradcast or season sets may not have had this issue, I cannot confirm.

THANKS TO EVERYONE FOR ALL YOUR HELP! Funimations used seperate tapes for the music, sound effects, and voices and this is why if you line up the music, then voices/sound effects will be off or vice versa.

skipping the gops with the ">" key and setting it "[" and then saving per episode from the .VOB files

mux and then go :) sorry if I was a little trouble, I could just tell on most episodes that there was a sync issue (more on some episodes than others), but atleast now I know why. I will just try my best to ignore the errors because there is literally nothing that can be done about it.

Edit: I have found that most episodes have a small overall audio alignment issue in one direction in addition to the sync discrepancy between the music & sound effects.
The Faulconer track used different music alltogether, so I have not been worrying about trying to correct sync based on music, only on sound effects and voice.
I have noticed with the discrepancy I have fixed in the audio alignment that when I adjust the delay to correct the alignment for soundeffects that both the lips and music are coming back into alignment.
I think the issue discussed on kanzenshuu about the sync issue between music and effect is only relevant to the tracks that have the original japanese music.
On those tracks I would prefer sound effect sync over music sync, you decide. but the track I am working on at the moment is the Faulconer track anyway.