How to fix speedup with AviSynth?

[VideoFanatic]

I've recently decided to re-encode this video where the 1st and 3rd part need the audio and video slowed down but the middle part is fine. Sorry I had to upload the whole thing so you can see the problem, I tried to make it small so it's around 250MB. You don't even need to encode the video to see the problem. You can just load the script and save the WAV file and if you look at 16 minutes - 21 mins 57 secs you can hear that the audio for the middle part was slowed down.

I set the trims so that the 1st and 3rd part of the video is slowed down and the middle part is left alone. Everything is fixed apart from the middle part which was fine to begin with but now it plays slower. What did I do wrong in my script?

vid=Mpeg2Source("E:\Fast Video (ECW Oct 19 1999).d2v", CPU=6)
aud=NicAC3Source("E:\Fast Video (ECW Oct 19 1999) T80 2_0ch 128Kbps DELAY 0ms.ac3")
AudioDub(vid,aud)
Main=last

Main
Trim(0,28787) # This part needs the audio and video slowed down
AssumeFPS(25) AssumeTFF() 
QTGMC(Preset="Super Fast", sharpness=0.7) 
source=last  
super = source.MSuper(pel=2)  
backward_vec = MAnalyse(super, overlap=4, isb = true, search=3)  
forward_vec = MAnalyse(super, overlap=4, isb = false, search=3)  
source.MFlowFps(super, backward_vec, forward_vec, blend=false, num=60000, den=1001) 
AssumeTFF().Vinverse().RemoveSpots().AddGrainC(var=1.0,uvar=1.0).SeparateFields().SelectEvery(4, 0, 3).Weave()
TimeStretch(rate=83.41666667) 
A=last

Main
Trim(28788,39486)  # This is 16 minutes - 21 mins 57 secs. It does NOT need to be slowed down.
AssumeTFF() QTGMC(Preset="Super Fast", sharpness=0.7)
Vinverse() RemoveSpots() AddGrainC(var=1.0,uvar=1.0) SeparateFields() SelectEvery(4,0,3) Weave() 
B=last

Main
Trim(39487,74377) # This part needs the audio and video slowed down
AssumeFPS(25) AssumeTFF() 
QTGMC(Preset="Super Fast", sharpness=0.7) 
source=last  
super = source.MSuper(pel=2)  
backward_vec = MAnalyse(super, overlap=4, isb = true, search=3)  
forward_vec = MAnalyse(super, overlap=4, isb = false, search=3)  
source.MFlowFps(super, backward_vec, forward_vec, blend=false, num=60000, den=1001) 
AssumeTFF().Vinverse().RemoveSpots().AddGrainC(var=1.0,uvar=1.0).SeparateFields().SelectEvery(4, 0, 3).Weave()
TimeStretch(rate=83.41666667)
C=last

A++B++C

SoundOut() # Once you've saved the audio you can comment this line out to prevent it asking you to save the audio every time you refresh the script in AvsPmod

[jagabo]

Looks like someone sped a 25 fps PAL video up to 29.97 fps.

[VideoFanatic]

The programme only aired in NTSC regions so I don't think it was originally a PAL video. Any suggestions on how to fix it though?

[jagabo]

Slow it down?

[VideoFanatic]

I know how to convert NTSC to PAL so should I just do that? What about the audio though? Won't the audio be out of sync with the video then? How do I slow down the audio?

[poisondeathray]

I know how to convert NTSC to PAL so should I just do that? What about the audio though? Won't the audio be out of sync with the video then? How do I slow down the audio?

There are many ways .

You can patch the elementary video with restream to 25 fps if you don't want to re-encode the video . Or if you want to make it completely PAL , resize to 720x576 use AssumeFPS(25), and re-encode (be careful about interlaced resizing method if you do)

To slow down audio, you can use various audio editors e.g. audacity, timestretch in avisynth with soundout

TimeStretch(rate=83.41666667)
Soundout()

Timestretch is expressed as a percentage, so 25/(30000/1001) * 100 =~ 83.41666667

(but it won't be NTSC DVD compliant @ 720x480 , 25 frames per second)

[VideoFanatic]

The programme only aired in NTSC regions so I don't understand how it's playing so fast in the first place. The current 29.97 frame rate should be correct yet it's playing too fast. I can't convert to PAL because you can't have a PAL video on the same disc that has NTSC videos on it.

I'll try your audio solution but is there any way to fix the video without converting to PAL?

[poisondeathray]

The programme only aired in NTSC regions so I don't understand how it's playing so fast in the first place. The current 29.97 frame rate should be correct yet it's playing too fast. I can't convert to PAL because you can't have a PAL video on the same disc that has NTSC videos on it.

I'll try your audio solution but is there any way to fix the video without converting to PAL?

It's playing fast because it was PAL at some point in it's life as jagabo mentioned. If you slow it down (both video & audio ) to 25 FPS it sounds and looks about right.

DGPulldown works only with progressive material, one option is to re-encode the video as 720x480 progressive, 25fps , and use DGPulldown to make it NTSC DVD compliant (output NTSC 29.97 signal using flags) , and use the slowed down audio . But you lose 1/2 the motion samples (will look choppy) when you deinterlace

I can't think of any good way to do this on a NTSC DVD

[poisondeathray]

Maybe another approach would be to use the slowed down audio , interpolate the frames

For the video, AssumeFPS(25), Bob Deinterlace to 50p (eg using QTGMC) , MVTools or simlar tools like svpflow or interframe to interpolate to 59.94p, Reinterlace as 59.94 fields per second

But interpolation often leads to edge morphing artifacts. You can use blends (convertfps) or dupes (changefps), but blends are blurry, dupes are choppy

You can expermiment and see which option you like better

[VideoFanatic]

Or could I just convert to PAL to fix the video. And fix the audio seperately. Then convert the video back to NTSC? I've converted from NTSC to PAL before and I didn't need to do anything to the audio so it should work.

[poisondeathray]

Or could I just convert to PAL to fix the video. And fix the audio seperately. Then convert the video back to NTSC? I've converted from NTSC to PAL before and I didn't need to do anything to the audio so it should work.

But that's the problem, converting back to NTSC. How would you convert back to NTSC? I listed the 3 methods above. This isn't a typical "NTSC speedup". You can't just speed it up, or you're left with the same thing as the orginal

PAL <=> NTSC usually refer to 25p vs. 24p (ie. 23.976p) the percentage difference is small . It's usually done with a simple slowdown, or speedup for the reverse

In contrast, 25 vs. 29.97 is a large % difference. That's why you can hear "chipmunk" voices . Someone screwed up this one.

[jagabo]

I've converted from NTSC to PAL before and I didn't need to do anything to the audio so it should work.

No. In this case you are going to slow the video down. So the running time will be longer. You can use Audacity to change the running time of the audio. Actually, I think AviSynth has built in audio features for that too. But I haven't used them. See ResampleAudio(), AssumeSampleRate(), TimeStretch(), etc.

[poisondeathray]

Here is an example using the interpolate method (mflowfps) for the video , reinterlaced, audio done with timestretch() , mentioned earlier

(I didn't bother doing any cleanup, it's just to demonstrate the motion , sync, and audio pitch/tempo )

[manono]

Nice job, pdr.

[VideoFanatic]

Thanks poisondeathray, the video sample you posted looks fine and you've improved the audio as well. Could you please share the script you used to do all that?

Also, do you think that's the best audio we can hope for or can it be improved because it still sounds a little weird.

[poisondeathray]

Thanks poisondeathray, the video sample you posted looks fine and you've improved the audio as well. Could you please share the script you used to do all that?

Also, do you think that's the best audio we can hope for or can it be improved because it still sounds a little weird.

The audio was done with timestretch described in post #6 above. You can play with the other parameters, you might get a better result. I used the default settings for seekwindow, overlap etc...
http://avisynth.org/mediawiki/TimeStretch

Or using a dedicated audio editor will probably give you better results.



The video was done by bob deinterlacing to 50p , interpolating to 59.94p using mflowfps, then reinterlacing
You can clean up noise and fix the colors too if you want

MPEG2Source()
AssumeFPS(25)
AssumeTFF()
Crop(8,0,0,0)
QTGMC(preset="faster", sharpness=0.7)

source=last
super = source.MSuper(pel=2)
backward_vec = MAnalyse(super, overlap=4, isb = true, search=3)
forward_vec = MAnalyse(super, overlap=4, isb = false, search=3)
source.MFlowFps(super, backward_vec, forward_vec, blend=false, num=60000, den=1001)

AddBorders(4,0,4,0)
AssumeTFF().SeparateFields().SelectEvery(4, 0, 3).Weave()

[VideoFanatic]

My video has different parts that play fine and other have speedup. How would I go about using your script on just the parts that need slowed down?

I normally do this to have different settings on different parts of a video:

filtered=Tweak(Hue=-0, Sat=1.0, Bright=0, Cont=1.0, Coring=False) ReplaceFramesSimple(last,filtered,Mappings="[0 6151]") 

filtered=Tweak(Hue=-10, Sat=1.0, Bright=0, Cont=1.0, Coring=False) ReplaceFramesSimple(last,filtered,Mappings="[6152 80000]")

[poisondeathray]

You can divide the different parts up using trim() , apply the various filters to sections , then join them with aligned splice

[VideoFanatic]

What's aligned splice?

[poisondeathray]

What's aligned splice?

Iit's appending segments using "++" . You're less likely to have audio problems than unaligned splice "+"

http://avisynth.org/mediawiki/AlignedSplice


e.g

a = avisource()
b = avisource()

a ++ b

[VideoFanatic]

Why do I need to use the above method? Doesn't trim join the segments anyway such as this:

trim(0, 20512). Tweak(Hue=-0, Sat=1.0, Bright=0, Cont=1.0, Coring=False) ++trim(20512,78229). Tweak(Hue=-0, Sat=1.0, Bright=0, Cont=1.0, Coring=False)

[poisondeathray]

Why do I need to use the above method? Doesn't trim join the segments anyway such as this:

trim(0, 20512). Tweak(Hue=-0, Sat=1.0, Bright=0, Cont=1.0, Coring=False) ++trim(20512,78229). Tweak(Hue=-0, Sat=1.0, Bright=0, Cont=1.0, Coring=False)

Trim doesn't join, the "++" joins segments


You had a thread about this already. I suggest you go back and re-read that. There are many different ways you can organize your scripts . If there are not too many segments, I like breaking them out into separate sections and giving them labels/names. The danger of using trim() is that it refers to "last", but I sometimes lose track of what "last" meant. You end up mixing filters on the wrong section or double filtering




The way I like doing it is something like this

Main=WhateverSource()


Main
Trim(0,100)
Filter1()
Filter2()
A=last


Main
Trim(500,600)
Filter3()
Filter4()
B=last


Main
Trim(800,900)
Filter5()
Filter6()
C=last

A ++ B ++ C

I made up imaginary "trims" . "A" refers to 0-100 , "B" refers to 500-600, "C" refers to 800-900) . You can give them more descriptive names too, like "intro", or "stong_filtered", etc... instead of A,B,C

You used ReplaceFramesSimple in an earlier script, you can do it that way too .

[manono]

You end up mixing filters on the wrong section or double filtering

Or putting the pieces back together incorrectly:

trim(0, 20512). Tweak(Hue=-0, Sat=1.0, Bright=0, Cont=1.0, Coring=False) ++trim(20512,78229). Tweak(Hue=-0, Sat=1.0, Bright=0, Cont=1.0, Coring=False)

[jagabo]

ReplaceFramesSimple() won't work because he's changing the number of frames (speed) of one of the videos.

[poisondeathray]

ReplaceFramesSimple() won't work because he's changing the number of frames (speed) of one of the videos.

whoops you're right ! He's going to be interpolating / adding more frames to those sections

[jagabo]

What you want is something like:

src=AviSource()

p1 = Trim(0,10).Tweak(...)
p2 = Trim(11,20).Tweak(...)
p3 = Trim(21,30).Tweak(...)

p1++p2++p3

[VideoFanatic]

I also tried all the trim methods you mentioned but none of them worked. To recap, my video & audio is playing too fast and I need to slow it down. However a middle part of the video plays fine without problems. So I need to only slow down the fast parts. The below script works however I cannot get your solution to work where it only applies the fix to the fast parts. Can someone please give an example script where it only applies the fixes to specific sections?

Mpeg2Source("E:\2 = New\SpeedUp.d2v", CPU=6)

AssumeFPS(25)

AssumeTFF() 
QTGMC(Preset="Super Fast", sharpness=0.7) 

source=last 
super = source.MSuper(pel=2) 
backward_vec = MAnalyse(super, overlap=4, isb = true, search=3) 
forward_vec = MAnalyse(super, overlap=4, isb = false, search=3) 
source.MFlowFps(super, backward_vec, forward_vec, blend=false, num=60000, den=1001)

AssumeTFF().SeparateFields().SelectEvery(4, 0, 3).Weave()

[jagabo]

src=Mpeg2Source("E:\2 = New\SpeedUp.d2v", CPU=6)

part1 = Trim(0,1000)
part3 = Trim (2000,0)

#part2
Trim (1001,1999)
AssumeFPS(25)
AssumeTFF() 
QTGMC(Preset="Super Fast", sharpness=0.7) 
super = MSuper(pel=2) 
backward_vec = MAnalyse(super, overlap=4, isb = true, search=3) 
forward_vec = MAnalyse(super, overlap=4, isb = false, search=3) 
source.MFlowFps(super, backward_vec, forward_vec, blend=false, num=60000, den=1001)
AssumeTFF().SeparateFields().SelectEvery(4, 0, 3).Weave()
 
part1++part2++part3

[poisondeathray]

Just break it out .

Main=Mpeg2Source()

Main
Trim(0,100)
A=last

Main
Trim(101,200)
AssumeFPS(25) 
AssumeTFF()  
QTGMC(Preset="Super Fast", sharpness=0.7) 
source=last  
super = source.MSuper(pel=2)  
backward_vec = MAnalyse(super, overlap=4, isb = true, search=3)  
forward_vec = MAnalyse(super, overlap=4, isb = false, search=3)  
source.MFlowFps(super, backward_vec, forward_vec, blend=false, num=60000, den=1001) 
AssumeTFF().SeparateFields().SelectEvery(4, 0, 3).Weave()
B=last

Main
Trim(201,300)
C=last

A++B++C

Change the trim numbers to the correct numbers corresponding to the sections

[jagabo]

Oh, I just realized the ends (part1 and part3 in my example) need the frame rate change, not the middle (part2). I'll leave it up to the OP to adjust the code.