IPB

Welcome Guest ( Log In | Register )

3 Pages V  < 1 2 3 >  
Reply to this topicStart new topic
Adobe Audition Stretching, Precise Stretching.
2Bdecided
post Jun 13 2013, 14:00
Post #26


ReplayGain developer


Group: Developer
Posts: 5058
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



QUOTE (Glenn Gundlach @ Jun 8 2013, 19:30) *
Someone mentioned the speed difference between PAL and NTSC. That peed difference is 25/23.976 - about 4.17% - more than 2/3 of a semitone. Your amount of change is 13900/13899 - less than 1/10 of the difference between NTSC B/W and NTSC color. Before pitch correction was practical the Europeans simply listened to American video 4% fast.
No, we listen to films fast: 24p sped up to 25p-in-50i. This is almost universal on TV and DVD (sometimes with pitch correction, especially on DVD, but often without), but not on BluRay (which is almost always native 24p). However, video converted from NTSC to PAL (60i to 50i) keeps its duration exactly.

Film transferred to NTSC video and then converted to PAL in the normal way also keeps its duration exactly (and has its picture quality wrecked!). This isn't common in the UK (if they show an excerpt from a new movie trailer on the news you'll sometimes see this, but otherwise I've never seen it), but in some markets and with some content types (especially anime), it's very common that a film is telecined to NTSC 24p-in-60i, field blend converted to PAL 50i, and then watched in that format - this gives you the correct audio duration and pitch, but lousy picture quality.

Cheers,
David.
Go to the top of the page
+Quote Post
zerowalker
post Jun 13 2013, 15:01
Post #27





Group: Members
Posts: 266
Joined: 6-August 11
Member No.: 92828



QUOTE (2Bdecided @ Jun 13 2013, 15:00) *
QUOTE (Glenn Gundlach @ Jun 8 2013, 19:30) *
Someone mentioned the speed difference between PAL and NTSC. That peed difference is 25/23.976 - about 4.17% - more than 2/3 of a semitone. Your amount of change is 13900/13899 - less than 1/10 of the difference between NTSC B/W and NTSC color. Before pitch correction was practical the Europeans simply listened to American video 4% fast.
No, we listen to films fast: 24p sped up to 25p-in-50i. This is almost universal on TV and DVD (sometimes with pitch correction, especially on DVD, but often without), but not on BluRay (which is almost always native 24p). However, video converted from NTSC to PAL (60i to 50i) keeps its duration exactly.

Film transferred to NTSC video and then converted to PAL in the normal way also keeps its duration exactly (and has its picture quality wrecked!). This isn't common in the UK (if they show an excerpt from a new movie trailer on the news you'll sometimes see this, but otherwise I've never seen it), but in some markets and with some content types (especially anime), it's very common that a film is telecined to NTSC 24p-in-60i, field blend converted to PAL 50i, and then watched in that format - this gives you the correct audio duration and pitch, but lousy picture quality.

Cheers,
David.



Arenīt all NTSC movies sped up to? (24 to 23.976), but thatīs pretty negligible though.

But so, if a movie is 60i, and converted to 50i, we keep the same, either skipping or blending frames?

Quite funny how PAL is either to fast, or to slow compared to NTSC depending on material.

And when you say NTSC film, do you mean 29.97?, cause when thatīs used in PAL, we either change it to 23.976 and speed it up, or we skip blend frames, making it "identical" to the NTSC version.

And as you say, TV are pretty much never pitch corrected, at least Analogue, i donīt think it ever was (may be wrong), and DVD is as you say, very rare, But it does exist.

The sad fact is though, that when dubbing is done in PAL for normal movies and series, we dub with 25fps, meaning we are dubbing to the sped up version, so we always get the background music etc, wrong, and it canīt be fixed.
However, for movies that have been shown at Cinema (24fps) we dub at that speed. So only material that have been shown at the Cinema has been dubbed correctly, which is very sad really.
Go to the top of the page
+Quote Post
zerowalker
post Sep 5 2013, 20:43
Post #28





Group: Members
Posts: 266
Joined: 6-August 11
Member No.: 92828



Can someone point me to the highest quality settings when using SOX and speed (for example, speed 0.96)?

Cause i noticed i get some noise currently, and i have been using

-rate -v speed 0.96

So it seems i must add dither, but if i just add dither, there is still noise
If i add dither -a it goes away, at least on silent parts.

It also goes away if i use -p 24.

So i am a bit confused.

The sound i am working with is 16bit and should end at 16bit after the conversion.

This post has been edited by zerowalker: Sep 5 2013, 20:43
Go to the top of the page
+Quote Post
Dynamic
post Sep 7 2013, 16:01
Post #29





Group: Members
Posts: 795
Joined: 17-September 06
Member No.: 35307



I'm a bit confused. I think there's a

rate

and a

--rate

but not a

-rate

option. I presume you're specifying rate to choose maximum quality (rate -v), though for 16-bit, it appears that rate -h is recommended, with -v intended for 24-bit. You could try rate -h instead to modify the quality of resampling that speed itself will default to.


--------------------
Dynamic – the artist formerly known as DickD
Go to the top of the page
+Quote Post
bandpass
post Sep 7 2013, 17:53
Post #30





Group: Members
Posts: 326
Joined: 3-August 08
From: UK
Member No.: 56644



The default rate parameters (giving –120dB) should be fine for 16-bit.

One possible source of noise is dither. Here's 3 different dither options: default (TPDF), none, and shaped:

CODE
sox -D -n -b 16 in.wav synth 8 sine 0:24k gain -1
sox    in.wav out1.wav speed .96
sox -D in.wav out2.wav speed .96
sox    in.wav out3.wav speed .96 dither -s

sox -M in.wav out[123].wav -n spectrogram -wk
display spectrogram.png &

(using linux syntax in the last two commands).

The other is clipping, so e.g.
CODE
sox -G in.wav out.wav speed .96
Go to the top of the page
+Quote Post
zerowalker
post Sep 7 2013, 21:29
Post #31





Group: Members
Posts: 266
Joined: 6-August 11
Member No.: 92828



QUOTE (bandpass @ Sep 7 2013, 18:53) *
The default rate parameters (giving –120dB) should be fine for 16-bit.

One possible source of noise is dither. Here's 3 different dither options: default (TPDF), none, and shaped:

CODE
sox -D -n -b 16 in.wav synth 8 sine 0:24k gain -1
sox    in.wav out1.wav speed .96
sox -D in.wav out2.wav speed .96
sox    in.wav out3.wav speed .96 dither -s

sox -M in.wav out[123].wav -n spectrogram -wk
display spectrogram.png &

(using linux syntax in the last two commands).

The other is clipping, so e.g.
CODE
sox -G in.wav out.wav speed .96


From my tests, Disabling Dither (-D) removes the noise. But i donīt really understand, why is it enabled if it adds noise, i thought dither was there to prevent the noise?
Go to the top of the page
+Quote Post
zerowalker
post Sep 7 2013, 21:33
Post #32





Group: Members
Posts: 266
Joined: 6-August 11
Member No.: 92828



QUOTE (Dynamic @ Sep 7 2013, 17:01) *
I'm a bit confused. I think there's a

rate

and a

--rate

but not a

-rate

option. I presume you're specifying rate to choose maximum quality (rate -v), though for 16-bit, it appears that rate -h is recommended, with -v intended for 24-bit. You could try rate -h instead to modify the quality of resampling that speed itself will default to.


Well that makes me confused as well. I thought, rate was rate, no matter if it is --rate or -rate etc.

But as you say, i am only using it to increase the quality of the resampling (as i guess thatīs used when changing speed).
It doesnīt make any difference if i change it, to the noise that is, the noise comes from Dithering which is a bit confusing for me.

But there shouldnīt be anything bad with choosing -v even if itīs 16 bit. I think they recommend it for 24bit, as itīs overkill, pretty much like placebo settings on stuff. But as it goes very fast anyway, i donīt see any reason to not use the best settings possible.
Go to the top of the page
+Quote Post
Wombat
post Sep 7 2013, 21:34
Post #33





Group: Members
Posts: 977
Joined: 7-October 01
Member No.: 235



Actualy dither IS noise. Do we talk about some noise you see in some zoomed spectral view and get annoyed or can you hear it? If so please provide a sample where we can hear the noise Sox adds.
Go to the top of the page
+Quote Post
lvqcl
post Sep 7 2013, 21:34
Post #34





Group: Developer
Posts: 3325
Joined: 2-December 07
Member No.: 49183



QUOTE (zerowalker @ Sep 8 2013, 00:29) *
From my tests, Disabling Dither (-D) removes the noise. But i donīt really understand, why is it enabled if it adds noise, i thought dither was there to prevent the noise?

No, dither adds noise to prevent distortion.

This post has been edited by lvqcl: Sep 7 2013, 21:36
Go to the top of the page
+Quote Post
zerowalker
post Sep 7 2013, 21:38
Post #35





Group: Members
Posts: 266
Joined: 6-August 11
Member No.: 92828



QUOTE (Wombat @ Sep 7 2013, 22:34) *
Actualy dither IS noise. Do we talk about some noise you see in some zoomed spectral view and get annoyed or can you hear it? If so please provide a sample where we can hear the noise Sox adds.


Ah now that you mention it, itīs there to hide more apparent noise and stuff like that. But i still think itīs wierd that it only adds noise in this case.

And havenīt looked at it in a spectral view so canīt speak for that. But itīs hearable, you canīt hear it, at least not well, if there is normal music or sound, as the noise getīs swallowed.
But if itīs very low sound, or silence, the noise is very apparent.

So the noise floor is pretty high, but not That high.

But if i disable dither, there is no noise at all (at least not audible).
Go to the top of the page
+Quote Post
zerowalker
post Sep 7 2013, 21:41
Post #36





Group: Members
Posts: 266
Joined: 6-August 11
Member No.: 92828



QUOTE (lvqcl @ Sep 7 2013, 22:34) *
QUOTE (zerowalker @ Sep 8 2013, 00:29) *
From my tests, Disabling Dither (-D) removes the noise. But i donīt really understand, why is it enabled if it adds noise, i thought dither was there to prevent the noise?

No, dither adds noise to prevent distortion.


Oh, well, am i supposed to use dithering for what i am doing or what;S?
I was on the impression that using dithering, is often a good thing, and shaped dithering is often better to hide noise at certain hertz.

I only want to change the speed at the highest quality possible with sox.
Go to the top of the page
+Quote Post
Wombat
post Sep 7 2013, 21:52
Post #37





Group: Members
Posts: 977
Joined: 7-October 01
Member No.: 235



Still i wonder how silent your music is to be able to hear the sox dither noise. It must be so silent you better go with 24bit or try to maximize its volume. Upping the volume on silent parts only to hear the noise makes no sense. Dither is designed to be not audibke at normal listening levels.
Go to the top of the page
+Quote Post
zerowalker
post Sep 7 2013, 21:59
Post #38





Group: Members
Posts: 266
Joined: 6-August 11
Member No.: 92828



QUOTE (Wombat @ Sep 7 2013, 22:52) *
Still i wonder how silent your music is to be able to hear the sox dither noise. It must be so silent you better go with 24bit or try to maximize its volume. Upping the volume on silent parts only to hear the noise makes no sense. Dither is designed to be not audibke at normal listening levels.


Not like that, the thing is. Itīs not pure music, itīs a Movie.

But the noise is clearly hearable if there is silence. I hear it very clearly at normal volume, the same volume i have to listen to the Movie itself.

But if there is a Movie Sound, that will swallow the noise, but it should still be there. Not really a problem as itīs not audible. But it shouldnīt be there, i mean, no dither makes silence, well silent.
With dithering, silence becomes noise, or well the nosie is everywhere, so the noise floor is at a normal range, but not that high that it overcomes other sounds.

Hope you get what i mean.
Go to the top of the page
+Quote Post
Wombat
post Sep 7 2013, 22:07
Post #39





Group: Members
Posts: 977
Joined: 7-October 01
Member No.: 235



Still i wonder. Flat dither can add a maximum of 6dB noise to your so called silence, leave alone noise shaped dither. If that is a problem you should have already problems without any dither.

This post has been edited by Wombat: Sep 7 2013, 22:09
Go to the top of the page
+Quote Post
zerowalker
post Sep 7 2013, 22:17
Post #40





Group: Members
Posts: 266
Joined: 6-August 11
Member No.: 92828



QUOTE (Wombat @ Sep 7 2013, 23:07) *
Still i wonder. Flat dither can add a maximum of 6dB noise to your so called silence, leave alone noise shaped dither. If that is a problem you should have already problems without any dither.


Well thatīs weird, not that good with Decibels, but should i be able to hear 6db normally, or would i need to increase the volume very high?

Cause i am not using shaped (have tried that, sounds the same i think), and i have no problems before, silence is silent.
Go to the top of the page
+Quote Post
Wombat
post Sep 7 2013, 22:30
Post #41





Group: Members
Posts: 977
Joined: 7-October 01
Member No.: 235



I doubt a movie audio stream has silence with absolute zero. It should have its own noisefloor you simply increase a bit with dither. If this paticular movie has not you should at least have problems with other movies' noise. It makes no sense to me otherwise, good luck.
Go to the top of the page
+Quote Post
zerowalker
post Sep 7 2013, 23:16
Post #42





Group: Members
Posts: 266
Joined: 6-August 11
Member No.: 92828



QUOTE (Wombat @ Sep 7 2013, 23:30) *
I doubt a movie audio stream has silence with absolute zero. It should have its own noisefloor you simply increase a bit with dither. If this paticular movie has not you should at least have problems with other movies' noise. It makes no sense to me otherwise, good luck.


Well probably, i donīt think itīs completely silent, but itīs not audible, even with increased volume, so itīs extremely low then.

But i still donīt know. Am i supposed to use Dither or Not?
Didnīt think it would be this problematic;S
Go to the top of the page
+Quote Post
Glenn Gundlach
post Sep 8 2013, 06:18
Post #43





Group: Members
Posts: 364
Joined: 19-April 08
From: LA
Member No.: 52914



I'm sorry. I was going off on a tangent about video standards conversion. How can I delete a post?



This post has been edited by Glenn Gundlach: Sep 8 2013, 06:28
Go to the top of the page
+Quote Post
[JAZ]
post Sep 8 2013, 10:47
Post #44





Group: Members
Posts: 1751
Joined: 24-June 02
From: Catalunya(Spain)
Member No.: 2383



@zerowalker:

In your scenario, shaped dither will make a difference, because you most probably are hearing the noise between 1...4Khz, while shaping it would move it up to 10Khz+.

Movies are usually of lower volume (less dynamic compression, and more regulated), so we could perfectly be in a scenario equivalent to 12 or 13bits.
Also, I don't know what you use to playback, but there could be some DRC or envelope follower in your equipment that in fact, increases slightly the volume on parts with less volume.
Also, you don't mention if you listen with headphones, or in a 5.1 setup. Headphones are definitely going to make noise more audible, because they would soften other noises around.


Very quick (but I hope, "to the point") explanation of dither:

1)Scenario:
-Integer digital formats have a finite number of values and cannot describe points inbetween.
-Signals are continuous, so a value at a specific time, can take any value.
-Sampling implies getting a value at periodic intervals (sample rate), and quantize that value (bit depth).
-Quantizing by truncation (round to zero) causes the so-called quantization distortion. Concretely, since the value is different than the original, the process is in fact adding a new signal into the original one. This signal is correlated to the original one, and changes according to the first. This makes it clearly visible in spectograms and audibly more annoying than simple noise.

2) Dither:
-Quantizing by dithering (rounding by adding a defined type of noise, white, triangular pdf, Gueiss...) causes dither noise. Again, the value is different than the original, and this also implies adding a new signal into the original one. But in this case, the signal is both, not correlated, and rather constant (in terms of variation during time). It is also very visible in spectograms, given enough bit depth, but being just noise, it is more pleasant that a distortion that matches the signal frequencies.
- Noise shaping is conceptually an equalizer that softens some frequencies to boost others. Noise shaping the dither noise previous to adding it to the original signal is implemented so that frequencies that humans are more sensible with have less noise, and those that were are less sensitive have more noise. When this shaped noise signal is added to the original, we still get the benefits of dithering (i.e. the added signal is not correlated), and we remove part of the problems of dithering (the added noise).


Edit:
Mmm.. that made me think... could we ride the dither noise by an envelope follower? So as to apply less noise on low volume (lower than maybe -40dBFS). Of course then we would have both, noise and quantization distortion, but it is a scenario to test what would be more pleasant.

This post has been edited by [JAZ]: Sep 8 2013, 11:04
Go to the top of the page
+Quote Post
zerowalker
post Sep 8 2013, 18:35
Post #45





Group: Members
Posts: 266
Joined: 6-August 11
Member No.: 92828



QUOTE ([JAZ] @ Sep 8 2013, 11:47) *

@zerowalker:

In your scenario, shaped dither will make a difference, because you most probably are hearing the noise between 1...4Khz, while shaping it would move it up to 10Khz+.

Movies are usually of lower volume (less dynamic compression, and more regulated), so we could perfectly be in a scenario equivalent to 12 or 13bits.
Also, I don't know what you use to playback, but there could be some DRC or envelope follower in your equipment that in fact, increases slightly the volume on parts with less volume.
Also, you don't mention if you listen with headphones, or in a 5.1 setup. Headphones are definitely going to make noise more audible, because they would soften other noises around.


Very quick (but I hope, "to the point") explanation of dither:

1)Scenario:
-Integer digital formats have a finite number of values and cannot describe points inbetween.
-Signals are continuous, so a value at a specific time, can take any value.
-Sampling implies getting a value at periodic intervals (sample rate), and quantize that value (bit depth).
-Quantizing by truncation (round to zero) causes the so-called quantization distortion. Concretely, since the value is different than the original, the process is in fact adding a new signal into the original one. This signal is correlated to the original one, and changes according to the first. This makes it clearly visible in spectograms and audibly more annoying than simple noise.

2) Dither:
-Quantizing by dithering (rounding by adding a defined type of noise, white, triangular pdf, Gueiss...) causes dither noise. Again, the value is different than the original, and this also implies adding a new signal into the original one. But in this case, the signal is both, not correlated, and rather constant (in terms of variation during time). It is also very visible in spectograms, given enough bit depth, but being just noise, it is more pleasant that a distortion that matches the signal frequencies.
- Noise shaping is conceptually an equalizer that softens some frequencies to boost others. Noise shaping the dither noise previous to adding it to the original signal is implemented so that frequencies that humans are more sensible with have less noise, and those that were are less sensitive have more noise. When this shaped noise signal is added to the original, we still get the benefits of dithering (i.e. the added signal is not correlated), and we remove part of the problems of dithering (the added noise).


Edit:
Mmm.. that made me think... could we ride the dither noise by an envelope follower? So as to apply less noise on low volume (lower than maybe -40dBFS). Of course then we would have both, noise and quantization distortion, but it is a scenario to test what would be more pleasant.


Very true, Shaped should be more pleasant, though when i tried it didnīt seem any different.

I donīt think any Dynamic Compression is going on, there shouldnīt be any, as i am working with Flac from AC3, and i never use DRC with AC3 for that matter.

I am also using Headphones (HD 280Pro), a Sound Card (ZxR Creative) and a Headphone Amp (O2).

I have uploaded 2 clips , one with the dither noise, one without.

I am letting you guess which is which (please only listen, do not check with Spectral etc).

Both are identical, except for one difference. "input" -D "output" trim 0 5 rate -v speed 0.96 , one uses this, and the other uses the same but Not with "-D".
So, both are identical except for the dithering part.

I clearly hear the noise of one of them.

There is silence at the beginning, then the intro starts slowly.

This post has been edited by zerowalker: Sep 8 2013, 18:37
Go to the top of the page
+Quote Post
[JAZ]
post Sep 8 2013, 19:01
Post #46





Group: Members
Posts: 1751
Joined: 24-June 02
From: Catalunya(Spain)
Member No.: 2383



errm.. uploaded where?

And please, do not full quote. (you can edit your post within one hour of your initial post time. else, i guess a moderator will do it, and maybe warn you smile.gif)
Go to the top of the page
+Quote Post
zerowalker
post Sep 9 2013, 02:31
Post #47





Group: Members
Posts: 266
Joined: 6-August 11
Member No.: 92828



Damn forgot to link: http://www.sendspace.com/filegroup/T6XZOuvplNWQzBn6MHL1Zw

And yeah sorry, usually let the Reply does it thing, didnīt think about that the post was quite big, my bad.
Go to the top of the page
+Quote Post
Glenn Gundlach
post Sep 14 2013, 10:35
Post #48





Group: Members
Posts: 364
Joined: 19-April 08
From: LA
Member No.: 52914



QUOTE (zerowalker @ Jun 5 2013, 15:30) *
Is it possible to stretch audio by a very precise amount in Adobe Audition?

Currently, i am talking about 500ms over the entire duration.
so for example.

1:55:50:000
will become: 1:55:50:500.


OK I tried a little experiment with Audition 3. First I made a file of 6950 seconds of 1KHz tone of length 1:55.50.000. Then I 'stretched' it at a ratio of 100.0071948 (13900/13899). The resulting file is now 1:55.50.500.

Does that help?

Go to the top of the page
+Quote Post
[JAZ]
post Sep 14 2013, 13:04
Post #49





Group: Members
Posts: 1751
Joined: 24-June 02
From: Catalunya(Spain)
Member No.: 2383



Hi.
Sorry, i missed your reply.

I checked both files (listened in foobar2000 and checked spectra in Audacity).
With (stereo) headphones, playing at a moderate volume, i wasn't able to hear the noise.
Then i started to test, and I was able to start hearing the noise when adding +10dB to the test.flac file. But still, that noise was not as high as the noise which the signal already has starting at second 1.8.


So, i wouldn't consider the noise added by dithering a bigger problem than what it really is.
It might be interesting to do what I suggested above, about applying less noise when the peak volume is low, but in this file, the problem really is the use of 16bit integer for a quiet signal.
Go to the top of the page
+Quote Post
zerowalker
post Sep 15 2013, 02:44
Post #50





Group: Members
Posts: 266
Joined: 6-August 11
Member No.: 92828



QUOTE ([JAZ] @ Sep 14 2013, 14:04) *

Hi.
Sorry, i missed your reply.

I checked both files (listened in foobar2000 and checked spectra in Audacity).
With (stereo) headphones, playing at a moderate volume, i wasn't able to hear the noise.
Then i started to test, and I was able to start hearing the noise when adding +10dB to the test.flac file. But still, that noise was not as high as the noise which the signal already has starting at second 1.8.


So, i wouldn't consider the noise added by dithering a bigger problem than what it really is.
It might be interesting to do what I suggested above, about applying less noise when the peak volume is low, but in this file, the problem really is the use of 16bit integer for a quiet signal.


Ah good, well itīs precisely as you say, the noise doesnīt really bother the rest of the sound, as itīs to low.

But the thing is, why should i even had dithering on, if i get noise with it and it seems to sound the same?
Or will it prevent harmonic distortions, cause i havenīt really done any depth analysis of that.

Not sure how to make it apply different depending on the peak. But i would prefer not to use, Adaptive solutions, they usually have some problematics , as they arenīt bulletproof.

And itīs probably as you say, the 16bit that makes the noise. But 16bit itself shouldnīt have that noise, so i guess itīs only working with 16bit.

Shouldnīt Sox be working in Float and then dithering it down to 16bit, making it as good as possible?

As i know that working only as one plane as 16bit, will create much more audible rounding errors, compared to working in float and then truncate it to 16bit.
Go to the top of the page
+Quote Post

3 Pages V  < 1 2 3 >
Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 



RSS Lo-Fi Version Time is now: 23rd July 2014 - 14:23