Pitch shift for PAL<->NTSC film conversion

[Xesdeeni]

I'm sorry if this should have been in the "New A/V Formats" section, which was recommended by doom9 for standards conversion, but I thought it might be more appropriate in the audio section instead.

It appears that when converting film-based video between NTSC and PAL, we all tend toward just speeding up or slowing down the video and doing the same to the audio as well. In some cases, this seems OK, and it appears that all early analog conversions were done this way. I've also advocated this process myself many times, including on my (under construction) Standards Conversion page (http://www.geocities.com/xesdeeni2001/StandardsConversion/index.html).

But I've never actually had side-by-side PAL and NTSC versions of the same movie to compare, in order to listen to what the professionals do today. So, in doing some searching in reference to the above-mentioned page above, I've come across several references to pitch the pitch shifting that is done in modern conversions. Here are a couple:

http://www.disctronics.co.uk/technology/video/video_film.htm
http://isb.ri.ccf.org/biomch-l/archives/biomch-l-1995-05/00087.html

As far as I can tell, BeSweet doesn't support pitch shifting, and neither does AVISynth. The newest VirtualDub adds some audio processing, but the dox clearly state that "The pitch shifter sucks."

In the interest of improving the film conversions on the above-mentioned page, is there another freely available tool that does a reasonable job of pitch shifting by the ~4% needed for this type of conversion?

Xesdeeni

[sh0dan]

Timestretching sound is in general not something I'd recommend - It tends to create a slight echoing effect, and blurs out transients in the sound - making it much more dull.

Also there is much difference between NTSC and film. NTSC is ALWAYS 30fps video, film is mostly 24fps. When talking about NTSC I'll be talking about 30fps (interlaced) material.

In short I'd recommend:

NTSC -> PAL - Deinterlace to 60fps using framedoubling (smoothdeinterlace for AviSynth / 100fps method). Decimate material to 50fps, reinterlace. Same the other way around.

FILM -> NTSC - Telecine.

NTSC -> FILM - Inverse Telecine

FILM -> PAL - Speed Up

PAL -> FILM - Deinterlace to 50fps - decimate. Or deinterlace + slow down.

[Xesdeeni]

NTSC -> PAL - Deinterlace to 60fps using framedoubling (smoothdeinterlace for AviSynth / 100fps method). Decimate material to 50fps, reinterlace. Same the other way around.I'm somewhat familiar with this... :)

But, I guess I wasn't clear in what I was asking.

I'm talking about conversion of film on NTSC to film on PAL or vice-versa. Using the above process will convert the video without changing the tempo, and without modifying the sound. However, the quality of the transfer will suffer from exagerated motion artifacts.

When mastering film to video, the professionals do as you outlined:FILM -> NTSC - Telecine.
...
FILM -> PAL - Speed Up
So, as is generally recommended here on doom9 (and in my Standards Conversion page mentioned above), the process of converting film on NTSC to film on PAL (and vice-versa) combines the above processes with the other two you mentioned:NTSC -> FILM - Inverse Telecine
...
PAL -> FILM - Deinterlace to 50fps - decimate. Or deinterlace + slow down.Here "FILM" is an intermediate step, but the process closely mimics what the professionals do when mastering film to video.

For NTSC->PAL, you would inverse telecine ("NTSC -> FILM - Inverse Telecine") and then speed up the film ("FILM -> PAL - Speed Up").

For PAL->NTSC, you would deinterlace ("PAL -> FILM - Deinterlace...") and then slow down to NTSC ("...slow down." & "FILM -> NTSC - Telecine").

Anyway, the bottom line is that the sound is pitch shifted by speeding up or slowing it down by about 4%. The links above indicate that the professionals reverse the pitch shift resulting from the speed change by doing a pitch shift of their own.Timestretching sound is in general not something I'd recommend - It tends to create a slight echoing effect, and blurs out transients in the sound - making it much more dull.I think more people object to the pitch shift due to the speed change than will object to some "muddying" of the sound at the corrected pitch. Besides, the references above indicate that the professionals do this, and people can certainly skip the pitch shift step if they prefer.

But I'm looking for a non-commercial tool for this process in order to provide the option in my guides.

Xesdeeni

[Ookami]

Originally posted by Xesdeeni
In the interest of improving the film conversions on the above-mentioned page, is there another freely available tool that does a reasonable job of pitch shifting by the ~4% needed for this type of conversion?

Xesdeeni

IIRC, there were some tools used in the beginning of the DVD backup (~ AD 1999 :D)... If you mean, stretching the audio, that is.

But, I really don't remember what the quality was.

I can search my old soft archive if you want. Mail me (don't use the PM, as my inbox is always full, or almost full) or reply here.

Cheers,

Mijo.

[Xesdeeni]

I'll take a look at anything you might be able to find!

So far, I've only been able to find the new audio plugin "ratty pitch shift" in the newest VirtualDub. I tried it with some video with audio and it sounded reasonable (at least for a first pass). But VirtualDub doesn't like audio only, so I can't use it on an extracted WAV file. When I tried to load in an AVISynth script with the audio (plus video), VirtualDub crashed.

I've download the newest VirtualDub source with an eye toward porting the code to an AVISynth filter, but I'm sure it will be at least two weeks or so before I'll have that much time at one sitting.

Xesdeeni

[Ookami]

Delete this one, please.

[Ookami]

Hello Xesdeeni...

http://www.doom9.org/software.htm and download "Wave lenght adjust" :) It even has the source included!

And check the attachment for the very old Wave convert by Extreme.

And here you can find some tools (like Audacity...), also:

http://ookami.videoxone.de/elist/audio.html

Hopefully this will help you.

Thanks for your work.

Cheers,

Mijo.

[Xesdeeni]

The "Wave Length Adjust" link appears to be broken. But the description just looks like it is a resampler (and if this is the same one as linked by the other page, it's a BAD resampler at that). That can be accomplished with most applications (on my Standards Conversion page, I use AVISynth for simplicity). But this changes the length as well as the pitch. Instead, we need an application that will change just the pitch, without changing the length.

Audacity doesn't appear to have a pitch shift either, nor do any other audio application on the above page.

That's the problem I've been having. Free applications don't seem to have this feature, most probably because they require some serious math (FFT/IFFT, etc.). Many commercial products will do this, but most of us will be converting only a few movies, so we don't want to have to pay the premiums.

Xesdeeni

[Atamido]

How would you change the length without the pitch, or vice-versa? If you have a certain wave of a given length, then changing the length would change the pitch, correct?

[Ookami]

> The "Wave Length Adjust" link appears to be broken

No it isn't, I've downloaded it today.

As for the other things, hopefully you'll find the program you need.

[Xesdeeni]

Check out the utility referenced in http://forum.doom9.org/showthread.php?s=&threadid=49800

Xesdeeni

[ihgl]

I am just a newbie and my simple question is: Is there no possibility whatever to do some kind of "telecining" from 23,976 to 25 and that way make the video length in the new "prolonged" PAL suit the original audiostream?? Can no suitable frame pattern be applied??

[Xesdeeni]

Certainly, but there is always a tradeoff. When you convert 23.976 to 25 fps, you have to fill in missing frames/fields. With a high cost motion compensated converter, the missing frames are created based on where objects would have been if the video were shot at the target frame rate. The quality will be quite good. But for what we can afford (and what will work on our current crop of PCs), the missing frames/fields have to either be duplicates or blends. I prefer duplicates to the blending, but with that you will get a "bump" on smooth panning shots. Depending on the content, this may not even be a problem. With blending, you get a "strobing" effect which I think is more objectionable.

Anyway, take a look at this thread: http://www.dvdrhelp.com/forum/viewtopic.php?t=160323 (ignore the 16:9 references if your movie is not anamorphic).

Xesdeeni

[ihgl]

Originally posted by Xesdeeni
Certainly, but there is always a tradeoff.


Anyway, take a look at this thread: http://www.dvdrhelp.com/forum/viewtopic.php?t=160323 (ignore the 16:9 references if your movie is not anamorphic).

Xesdeeni

Thanks a lot for your answer - and all other very informative posts. This item is really interesting. I will try the link you supplied. ;-)

[kenma]

@ Xesdeeni
is there something new with it since ?

[Xesdeeni]

I'm sorry, no. I haven't had time to get back to WSOLA (new house, new baby on the way, etc.). There was a guy who contacted me about the source, but I haven't heard anything more from him. I'm hoping that someone with more time will eventually take a look, but not so far. Also, the next update to my Standards Conversion page (also delayed due to the above real-world issues) will remove WSOLA from the process. If I get time in the future to address WSOLA again and fix the problems, I'll re-add it.

Xesdeeni