AviSynth and Yadif mode=1 truncates audio

F1nkster · 15th December 2011, 01:56

Using AVISynth 2.60 Alpha 3 with DV Type 2 AVI files on Windows 7 32-bit. Script:

Load_StdCall_Plugin("C:\Program Files\AviSynth 2.5\plugins\yadif.dll")
AviSource(srcvid).AssumeBFF()
Yadif(mode=1)
AssumeFPS("ntsc_double")

I've used FFmpeg, mplayer, and WMP to play the script and see consistent results on all 3.

I get video with correct duration at double framerate (as it should be). However the audio is getting truncated before the end of the video. The audio and video are in sync, it is just that the audio is truncated -- that is, it goes silent before the end of the video. I am losing 0.9 seconds of audio for every 15 minutes of video, so that on a 70 minute video that I have, I am losing a little over 4 seconds at the end. However, the ratio of 0.9 seconds for every 15 minutes holds precisely true, which is to say, if I trim the video to 15 minutes, 0.9 seconds of audio will be missing. At 30 minutes, I lose 1.8 seconds, and so on. And this is happening for every DV AVI file I throw at it.

If I remove Yadif/AssumeFPS or use mode=0, the audio outputs at the correct length. Iow, no issue. However, I am unwilling to give up Yadif(mode=1). I have not tested other deinterlacers (for this problem), nor do I plan to. So I pressed on...

AssumeFPS is necessary to maintain audio sync. Without it, Yadif outputs rounded 59.94 and not 60000/1001. Not noticeable on shorter videos, but audio goes noticeably out of sync on longer videos (say 70 minutes, for example). Sync is spot-on with AssumeFPS. That said...

I tried other import filters (FFVideoSource/FFAudioSource, DSS). Same result (that is, the problem manifests).
Demuxed audio and used WAVSource followed by AudioDub. Same result.
Placed AudioDub after Yadif/AssumeFPS. Same result.
Set AssumeFPS's sync_audio parameter to true. Same result.
Tried ConvertFPS instead of AssumeFPS. Same result.

I finally found a workaround, but not a fix. I add this to the script:

last++BlankClip(last,Round((0.9*last.framecount)/900))

In other words, I add a blank clip at the ratio of 0.9 seconds for every 15 minutes of video. This gives me all my audio and in sync with the video, but with some blank frames at the end. Most of my videos are shorter, so the blank frames are not that noticeable.

I don't know if the issue is with Yadif or with Avisynth or with how the two are interacting. Clearly there is a problem.

If you view Info on the clip in AviSynth, it reports video and audio as the same length. It is only when playing the script back, or dumping audio with ffmpeg (as in "ffmpeg -i test.avs -vn -acodec copy test.wav") that the I observe the problem.

If anyone has seen this, has any insight, or possibly even a fix, please help! I did find another post from 2009 that was less detailed, but I believe it described the same problem. There was no resolution. I am hopeful with the extra detail above, and the right eyes looking at it, the problem can be identified.

IanB · 15th December 2011, 05:08

It looks like Yadif might be truncating the audio to match the video.

As a test, replace Yadif(1) with SelectEvery(1, 0,0). This will just return each unprocessed frame twice, so you can test the Audio sync logic of the script without the Yadif plugin, i.e. prove if the problem is in Yadif(1) or something more fundamental.

Also put Info() calls between each statement in turn (one at a time) to see how the Audio and Video duration, frame count and sample count are all adjusted by each statement.

And try moving the AssumeFPS line to before the Yadif(1) line, but make it AssumeFPS("ntsc") of course.

F1nkster · 15th December 2011, 18:10

Thanks for the reply. So I replaced Yadif(1) with SelectEvery(1,0,0), and got the same result. The audio is truncated by precisely the same amount.

I also used AssumeFPS("ntsc_video") prior to Yadif (and SelectEvery on a separate run), and this output 60000/1001 video, so I will stick with it.

Also, when I use Info at any point in the script, the duration of the video and audio always match. It is only when executing the script, and outputting the audio, that the issue is noticed.

Also, the fact that I demuxed audio, imported with WAVSource, and used AudioDub after Yadif, also reinforces that Yadif is not the root cause.

IanB · 15th December 2011, 22:16

Quote:

Originally Posted by F1nkster

Thanks for the reply. So I replaced Yadif(1) with SelectEvery(1,0,0), and got the same result. The audio is truncated by precisely the same amount.

So the problem is not in Yadif(1) but is something more fundamental.

Quote:

Also, when I use Info at any point in the script, the duration of the video and audio always match. It is only when executing the script, and outputting the audio, that the issue is noticed.

If you process/play the script with VirtualDub does it also have a problem.

Does VirtualDub's Information popup agree with an Info() display at the very end of the script?

Perhaps you can put up a small screen grab of the Info() display for the last frame.

sneaker_ger · 16th December 2011, 00:39

In NTSC we get 1 extra second for every 1000 seconds of original film footage. (1000s/60s)*0.9s=15 minutes
Coincidence?
What happens if you use the following:
1.) AssumeFPS(60)
2.) AssumeFPS(60000, 1001)
?

F1nkster · 16th December 2011, 02:57

So IanB's suggestion of running the script through VDubMod got me started on a path that has convinced me the problem is not with AviSynth or Yadif. It appears to be with FFmpeg.

I can use VDubMod to save the WAV file off the avs script and it outputs correctly with no truncation.

I also downloaded a little utility called wavi and played the avs script through that, outputting a wav file. Again, no truncation.

I guess I was giving FFmpeg the benefit of the doubt earlier. Still, I played the script through mplayer and it seemed to cut-off, but now I am wondering if it just wasn't keeping up and getting out of sync.

Anyhow, I tried setting the framerate with FFmpeg using the -r parameter. It made no difference. I cannot get FFmpeg to output correct audio. So much for FFmpeg, at least for encoding the audio.

I can use wavi and pipe it to faac, then mux with an h264 file out of ffmpeg. Or mencoder for that matter. But ffmpeg was so nice because I could run it against the avs script one time and output an h264 and aac file, then mux with MP4Box. Now I have to process the script twice, once for video, once for audio. That sucks.

Thanks for the replies on this!

IanB · 16th December 2011, 05:17

The FFMpeg libraries are very widely used as a general purpose input (even Avisynth has FFMpegSource() available), so if it has a problem truncating the audio with Avisynth input, anything that uses it will exhibit the problem.

Hopefully an FFMpeg expert can provide some insight.

Seems the brute force approach of adding 1 or more seconds of video and audio padding with "Last + BlankClip(Last, 30)" might be the easiest approach.

TheRyuu · 16th December 2011, 06:23

So to reproduce:

Any input file?
Load video+audio in avisynth.
Throw script at ffmpeg?

vampiredom · 16th December 2011, 07:45

Quote:

Originally Posted by sneaker_ger

In NTSC we get 1 extra second for every 1000 seconds of original film footage. (1000s/60s)*0.9s=15 minutes

Yeah, I was reading the thread and thinking the same kind of thing. I'd like to know if the problem persists without the AssumeFPS() line at all, or perhaps...

Code:

AssumeFPS(60000,1001, sync_audio=true)

** EDIT ** I reread and it looks like you tried that already. Very odd.

F1nkster · 16th December 2011, 13:33

In response to TheRyuu, doubling the framerate is another necessary piece to reproducing the problem.

If you just load audio+video, and change nothing, then throw at ffmpeg, it works fine.

But use a bob deinterlacer that doubles the framerate, or even just SelectEvery(1,0,0) as we tested above, then we have the problem.

Removing AssumeFPS does not affect the truncating issue. However, removing it leads to an audio sync issue that is noticeable only on longer clips of say 60 minutes or more (roughly). The truncation issue exists with or without AssumeFPS. It just becomes harder to troubleshoot because you think you're dealing with an audio sync issue, when you are in fact dealing with both an audio sync and a truncation issue at that point. Actually, the truth is that you have a framerate issue that manifests as an audio sync issue. Anyway...

I have only tested DV AVI files. I have some HDV mpeg files that I will be getting to.

Gavino · 16th December 2011, 13:46

What is your ffmpeg command line?
(Any special parameters that might affect the output?)

F1nkster · 16th December 2011, 18:39

The problem can be observed simply by using the following:

ffmpeg -i test.avs -vn -acodec copy test.wav

I also tried inserting -r 59.94, -r 59.9401, and -r 60 at different points (before/after test.avs) and it had no effect.

In actual use, I was using ffmpeg to encode to aac and observing the issue that way as well but it is unnecessary to go through all that.

IanB · 17th December 2011, 09:19

Yes it seem FFMpeg chops a small amount off the end.

I tried :-

Code:

ColorBars()
Trim(0, -15*60*30) # 15:00.9 @ 29.97fps
Info()

The Info display confirmed 15:00.9 duration for audio and video. A "ffmpeg -i test.avs -vn -acodec copy test.wav" claimed 15:00.9 source duration. But it only wrote 168855Kb of audio, it should have written 168919Kb.

The subsequent test.wav file was only 15:00.56 duration. Which is of course is 168855Kb.

F1nkster · 19th December 2011, 15:20

IanB, I'm interpreting your post as saying that FFmpeg is truncating audio with/without doubling the framerate. I guess it just becomes more noticeable when doubling the framerate.

Anyhow, I've change my workflow and scripts to use mencoder for video and Nero AAC for audio, preprocessed with AVISynth of course. I just have to encode the streams separately then mux with MP4Box. So be it. Everything is working correctly without having to add a blank clip to the end.

TheRyuu · 20th December 2011, 01:03

Quote:

Originally Posted by F1nkster

IanB, I'm interpreting your post as saying that FFmpeg is truncating audio with/without doubling the framerate. I guess it just becomes more noticeable when doubling the framerate.

Anyhow, I've change my workflow and scripts to use mencoder for video and Nero AAC for audio, preprocessed with AVISynth of course. I just have to encode the streams separately then mux with MP4Box. So be it. Everything is working correctly without having to add a blank clip to the end.

If you're going to do them both separate anyway why not just use x264 directly.

Also qt aac will likely result in better quality than nero.

F1nkster · 20th December 2011, 19:04

Thanks for the tips. So many options! I'll check out qt aac.

I've considered just using x264 directly. I'm familiar with mencoder so just went down that road for expedience.

This may be worth a thread unto itself, but is there an advantage to using x264 vs mencoder (or ffmpeg for that matter)?

I've searched on this question and haven't found any definitive answers, just people's preferences. I was left with the impression that the output should be the same since they are all using x264.

F1nkster · 20th December 2011, 21:59

So I did some quick testing with using x264.exe for encoding. In doing so, I discovered that many of the x264 options that I thought I was setting with mencoder were not actually working -- meaning when I looked at mediainfo on the output file, the encode settings that it shows were not what I had set. So perhaps this points to an issue with mencoder. Anyhow I have found it more expedient to just use x264 than to troubleshoot mencoder. Fwiw, I was using a mencoder.conf file to create an x264 profile with options, so maybe I did something wrong there. It's not worth the time figuring it out. x264 wins.

15th December 2011, 01:56	#1 \| Link
F1nkster Registered User Join Date: Sep 2011 Location: King of Prussia, PA Posts: 19	AviSynth and Yadif mode=1 truncates audio Using AVISynth 2.60 Alpha 3 with DV Type 2 AVI files on Windows 7 32-bit. Script: Load_StdCall_Plugin("C:\Program Files\AviSynth 2.5\plugins\yadif.dll") AviSource(srcvid).AssumeBFF() Yadif(mode=1) AssumeFPS("ntsc_double") I've used FFmpeg, mplayer, and WMP to play the script and see consistent results on all 3. I get video with correct duration at double framerate (as it should be). However the audio is getting truncated before the end of the video. The audio and video are in sync, it is just that the audio is truncated -- that is, it goes silent before the end of the video. I am losing 0.9 seconds of audio for every 15 minutes of video, so that on a 70 minute video that I have, I am losing a little over 4 seconds at the end. However, the ratio of 0.9 seconds for every 15 minutes holds precisely true, which is to say, if I trim the video to 15 minutes, 0.9 seconds of audio will be missing. At 30 minutes, I lose 1.8 seconds, and so on. And this is happening for every DV AVI file I throw at it. If I remove Yadif/AssumeFPS or use mode=0, the audio outputs at the correct length. Iow, no issue. However, I am unwilling to give up Yadif(mode=1). I have not tested other deinterlacers (for this problem), nor do I plan to. So I pressed on... AssumeFPS is necessary to maintain audio sync. Without it, Yadif outputs rounded 59.94 and not 60000/1001. Not noticeable on shorter videos, but audio goes noticeably out of sync on longer videos (say 70 minutes, for example). Sync is spot-on with AssumeFPS. That said... I tried other import filters (FFVideoSource/FFAudioSource, DSS). Same result (that is, the problem manifests). Demuxed audio and used WAVSource followed by AudioDub. Same result. Placed AudioDub after Yadif/AssumeFPS. Same result. Set AssumeFPS's sync_audio parameter to true. Same result. Tried ConvertFPS instead of AssumeFPS. Same result. I finally found a workaround, but not a fix. I add this to the script: last++BlankClip(last,Round((0.9*last.framecount)/900)) In other words, I add a blank clip at the ratio of 0.9 seconds for every 15 minutes of video. This gives me all my audio and in sync with the video, but with some blank frames at the end. Most of my videos are shorter, so the blank frames are not that noticeable. I don't know if the issue is with Yadif or with Avisynth or with how the two are interacting. Clearly there is a problem. If you view Info on the clip in AviSynth, it reports video and audio as the same length. It is only when playing the script back, or dumping audio with ffmpeg (as in "ffmpeg -i test.avs -vn -acodec copy test.wav") that the I observe the problem. If anyone has seen this, has any insight, or possibly even a fix, please help! I did find another post from 2009 that was less detailed, but I believe it described the same problem. There was no resolution. I am hopeful with the extra detail above, and the right eyes looking at it, the problem can be identified.

16th December 2011, 13:33	#10 \| Link
F1nkster Registered User Join Date: Sep 2011 Location: King of Prussia, PA Posts: 19	In response to TheRyuu, doubling the framerate is another necessary piece to reproducing the problem. If you just load audio+video, and change nothing, then throw at ffmpeg, it works fine. But use a bob deinterlacer that doubles the framerate, or even just SelectEvery(1,0,0) as we tested above, then we have the problem. Removing AssumeFPS does not affect the truncating issue. However, removing it leads to an audio sync issue that is noticeable only on longer clips of say 60 minutes or more (roughly). The truncation issue exists with or without AssumeFPS. It just becomes harder to troubleshoot because you think you're dealing with an audio sync issue, when you are in fact dealing with both an audio sync and a truncation issue at that point. Actually, the truth is that you have a framerate issue that manifests as an audio sync issue. Anyway... I have only tested DV AVI files. I have some HDV mpeg files that I will be getting to. Last edited by F1nkster; 16th December 2011 at 13:39.

16th December 2011, 13:46	#11 \| Link
Gavino Avisynth language lover Join Date: Dec 2007 Location: Spain Posts: 3,431	What is your ffmpeg command line? (Any special parameters that might affect the output?) __________________ GScript and GRunT - complex Avisynth scripting made easier

17th December 2011, 09:19	#13 \| Link
IanB Avisynth Developer Join Date: Jan 2003 Location: Melbourne, Australia Posts: 3,167	Yes it seem FFMpeg chops a small amount off the end. I tried :- Code: ColorBars() Trim(0, -156030) # 15:00.9 @ 29.97fps Info() The Info display confirmed 15:00.9 duration for audio and video. A "ffmpeg -i test.avs -vn -acodec copy test.wav" claimed 15:00.9 source duration. But it only wrote 168855Kb of audio, it should have written 168919Kb. The subsequent test.wav file was only 15:00.56 duration. Which is of course is 168855Kb.

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

15th December 2011, 05:08	#2 \| Link
IanB Avisynth Developer Join Date: Jan 2003 Location: Melbourne, Australia Posts: 3,167	It looks like Yadif might be truncating the audio to match the video. As a test, replace Yadif(1) with SelectEvery(1, 0,0). This will just return each unprocessed frame twice, so you can test the Audio sync logic of the script without the Yadif plugin, i.e. prove if the problem is in Yadif(1) or something more fundamental. Also put Info() calls between each statement in turn (one at a time) to see how the Audio and Video duration, frame count and sample count are all adjusted by each statement. And try moving the AssumeFPS line to before the Yadif(1) line, but make it AssumeFPS("ntsc") of course.

15th December 2011, 18:10	#3 \| Link
F1nkster Registered User Join Date: Sep 2011 Location: King of Prussia, PA Posts: 19	Thanks for the reply. So I replaced Yadif(1) with SelectEvery(1,0,0), and got the same result. The audio is truncated by precisely the same amount. I also used AssumeFPS("ntsc_video") prior to Yadif (and SelectEvery on a separate run), and this output 60000/1001 video, so I will stick with it. Also, when I use Info at any point in the script, the duration of the video and audio always match. It is only when executing the script, and outputting the audio, that the issue is noticed. Also, the fact that I demuxed audio, imported with WAVSource, and used AudioDub after Yadif, also reinforces that Yadif is not the root cause.

16th December 2011, 00:39	#5 \| Link
sneaker_ger Registered User Join Date: Dec 2002 Posts: 5,565	In NTSC we get 1 extra second for every 1000 seconds of original film footage. (1000s/60s)*0.9s=15 minutes Coincidence? What happens if you use the following: 1.) AssumeFPS(60) 2.) AssumeFPS(60000, 1001) ?

16th December 2011, 02:57	#6 \| Link
F1nkster Registered User Join Date: Sep 2011 Location: King of Prussia, PA Posts: 19	So IanB's suggestion of running the script through VDubMod got me started on a path that has convinced me the problem is not with AviSynth or Yadif. It appears to be with FFmpeg. I can use VDubMod to save the WAV file off the avs script and it outputs correctly with no truncation. I also downloaded a little utility called wavi and played the avs script through that, outputting a wav file. Again, no truncation. I guess I was giving FFmpeg the benefit of the doubt earlier. Still, I played the script through mplayer and it seemed to cut-off, but now I am wondering if it just wasn't keeping up and getting out of sync. Anyhow, I tried setting the framerate with FFmpeg using the -r parameter. It made no difference. I cannot get FFmpeg to output correct audio. So much for FFmpeg, at least for encoding the audio. I can use wavi and pipe it to faac, then mux with an h264 file out of ffmpeg. Or mencoder for that matter. But ffmpeg was so nice because I could run it against the avs script one time and output an h264 and aac file, then mux with MP4Box. Now I have to process the script twice, once for video, once for audio. That sucks. Thanks for the replies on this!

16th December 2011, 05:17	#7 \| Link
IanB Avisynth Developer Join Date: Jan 2003 Location: Melbourne, Australia Posts: 3,167	The FFMpeg libraries are very widely used as a general purpose input (even Avisynth has FFMpegSource() available), so if it has a problem truncating the audio with Avisynth input, anything that uses it will exhibit the problem. Hopefully an FFMpeg expert can provide some insight. Seems the brute force approach of adding 1 or more seconds of video and audio padding with "Last + BlankClip(Last, 30)" might be the easiest approach.

16th December 2011, 06:23	#8 \| Link
TheRyuu warpsharpened Join Date: Feb 2007 Posts: 787	So to reproduce: Any input file? Load video+audio in avisynth. Throw script at ffmpeg?

16th December 2011, 18:39	#12 \| Link
F1nkster Registered User Join Date: Sep 2011 Location: King of Prussia, PA Posts: 19	The problem can be observed simply by using the following: ffmpeg -i test.avs -vn -acodec copy test.wav I also tried inserting -r 59.94, -r 59.9401, and -r 60 at different points (before/after test.avs) and it had no effect. In actual use, I was using ffmpeg to encode to aac and observing the issue that way as well but it is unnecessary to go through all that.

19th December 2011, 15:20	#14 \| Link
F1nkster Registered User Join Date: Sep 2011 Location: King of Prussia, PA Posts: 19	IanB, I'm interpreting your post as saying that FFmpeg is truncating audio with/without doubling the framerate. I guess it just becomes more noticeable when doubling the framerate. Anyhow, I've change my workflow and scripts to use mencoder for video and Nero AAC for audio, preprocessed with AVISynth of course. I just have to encode the streams separately then mux with MP4Box. So be it. Everything is working correctly without having to add a blank clip to the end.

20th December 2011, 19:04	#16 \| Link
F1nkster Registered User Join Date: Sep 2011 Location: King of Prussia, PA Posts: 19	Thanks for the tips. So many options! I'll check out qt aac. I've considered just using x264 directly. I'm familiar with mencoder so just went down that road for expedience. This may be worth a thread unto itself, but is there an advantage to using x264 vs mencoder (or ffmpeg for that matter)? I've searched on this question and haven't found any definitive answers, just people's preferences. I was left with the impression that the output should be the same since they are all using x264.

20th December 2011, 21:59	#17 \| Link
F1nkster Registered User Join Date: Sep 2011 Location: King of Prussia, PA Posts: 19	So I did some quick testing with using x264.exe for encoding. In doing so, I discovered that many of the x264 options that I thought I was setting with mencoder were not actually working -- meaning when I looked at mediainfo on the output file, the encode settings that it shows were not what I had set. So perhaps this points to an issue with mencoder. Anyhow I have found it more expedient to just use x264 than to troubleshoot mencoder. Fwiw, I was using a mencoder.conf file to create an x264 profile with options, so maybe I did something wrong there. It's not worth the time figuring it out. x264 wins.