IPB

Welcome Guest ( Log In | Register )

 
Reply to this topicStart new topic
help/opinons on multiple speed change quality - test
bomber1978
post Sep 24 2013, 11:57
Post #1





Group: Members
Posts: 27
Joined: 24-September 13
Member No.: 110284



Hello,
I'm not sure if I can post this here, or if anyone will be interested to help/give their opinion, I guess it can just be ignored if you're not interested to help...
I have been testing various programmes "speed change" function quality , as I have been pretty seriously correcting analog cassette tape live recordings speed...
In testing the quality of the "speed change" function I took an original file off a CD and changed the speed 9 times (yes I know that's an extreme amount but I wanted to push the test to the limits!).
The original file is 7min:9secs.26667 (5 decimals is the most I can see, beyond that I don't know...), after changing the speed of the original file 9 times I returned it to the original length of 7min9secs.26667, I then created a mixdown of the original file and speed changed file, as the mixdown cancels out audio which is the same so you can hear any difference between the 2 files....
By ear doing ABX type tests I can not hear any difference between the original file and the 9 times speed changed file (I keep thinking I can but blind tests have proven me wrong many more times than I was right!). So I created the mixdown, however I'm not sure if the mixdown results are accurate, because the mixdown starts off at a low volume and gets louder as the track lengthens, so I think this could be because of time lag, I'm guessing it's because the file time length doesn't match somewhere after the 5th decimal spot which I can not see..., I know the difference the mixdown is showing could possibly be digital noise because of bad quality speed change, but the fact the mixdown starts off at a low volume and gets louder seems to indicate to me that it's more of a time lag problem (I really don't know for sure which is why I'm asking for opinion here!).
I am hoping someone with much better knowledge than me (probably most people here!) would be able to please take a listen to my files, and please let me know your opinion, that would be greatly appreciated! I'm sure there has to be some time lag, but more importantly I'm wondering if anyone can hear any real difference in quality between the original file and speed changed 9 file?, as the purpose of my test is to test the quality of the speed changed file against the original...

The following 3 files can be downloaded here:
ORIGINAL FILE:
http://www.sendspace.com/file/smrrdv

SPEED CHANGED 9 TIMES FILE:
http://www.sendspace.com/file/9idl1n

MIXDOWN of original file and speed changed 9 times file:
http://www.sendspace.com/file/6380gy





Go to the top of the page
+Quote Post
pdq
post Sep 24 2013, 12:48
Post #2





Group: Members
Posts: 3412
Joined: 1-September 05
From: SE Pennsylvania
Member No.: 24233



I'm sure this is obvious, but if you can't ABX the result of nine time changes, doesn't that tell you that you are worrying about nothing?
Go to the top of the page
+Quote Post
bomber1978
post Sep 24 2013, 13:19
Post #3





Group: Members
Posts: 27
Joined: 24-September 13
Member No.: 110284



QUOTE (pdq @ Sep 24 2013, 21:48) *
I'm sure this is obvious, but if you can't ABX the result of nine time changes, doesn't that tell you that you are worrying about nothing?


Not necessarily, I know I have a fairly good ear for audio, but I can only do these tests on my laptop, which currently doesn't have the greatest soundcard, I know I can hear detail on my stereo system which I can't hear on my laptop with headphones, so it's not really ideal for me to do this test on the laptop I'm currently using...
And I know there are people here who know what digital artefacts introduced by speed changing may sound like, whereas I really don't know what to listen for..., does it sound the same as the digital "underwater / metallic" noise that noise/hiss reduction introduces?, or does it sound like something else?, for example a friend thought they could hear some echo/reverb on the speed changed file?, but I can not hear that?, so maybe there's something there to be heard which I can not hear?, once something is pointed out to me I can usually hear it and learn about it, I'm also using this as a way to try and learn more, so if there's something there which is audible I'd like to know!
Plus like I said, I do hear a difference in the mixdown, however I don't know if that is because of drift, or if it's because of difference in quality? - which was part of my question?, and reason I posted asking for help & opinion...
So I'm really hoping someone can give an opinion who has the proper equipment / ear to give an answer...


This post has been edited by bomber1978: Sep 24 2013, 13:36
Go to the top of the page
+Quote Post
bandpass
post Sep 24 2013, 13:35
Post #4





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



Bear in mind that changing speed digitally is implemented as resampling, which is completely transparent with many tools (see e.g. http://src.infinitewave.ca and other threads here), so (with a few common sense provisos such as not shifting frequencies above nyquist) not being ABXable, on any equipment, is to be expected.
Go to the top of the page
+Quote Post
Arnold B. Kruege...
post Sep 24 2013, 13:37
Post #5





Group: Members
Posts: 3812
Joined: 29-October 08
From: USA, 48236
Member No.: 61311



There are at least 2 different ways to change the speed of a file. What methodology are these samples using?

This post has been edited by db1989: Sep 24 2013, 18:45
Reason for edit: deleting pointless full quote of OP
Go to the top of the page
+Quote Post
bomber1978
post Sep 24 2013, 13:40
Post #6





Group: Members
Posts: 27
Joined: 24-September 13
Member No.: 110284



QUOTE (bandpass @ Sep 24 2013, 22:35) *
Bear in mind that changing speed digitally is implemented as resampling, which is completely transparent with many tools (see e.g. http://src.infinitewave.ca and other threads here), so (with a few common sense provisos such as not shifting frequencies above nyquist) not being ABXable, on any equipment, is to be expected.


Thanks I'm really not very good with those types of graphs!, which is why I asked if anyone could hear "by ear" any differences in the 3 audio samples I posted in my original post!, yes I know speed changing is resampling, but I'm not exactly sure what you are telling me, about whether or not the difference in my example is audible?
Go to the top of the page
+Quote Post
bomber1978
post Sep 24 2013, 13:44
Post #7





Group: Members
Posts: 27
Joined: 24-September 13
Member No.: 110284



QUOTE (Arnold B. Krueger @ Sep 24 2013, 22:37) *
There are at least 2 different ways to change the speed of a file. What methodology are these samples using?
Hi,
Not sure if this is the correct explanation, but I made the speed changes by % eg. load in the file and make +3.0% change, then -2.5% change (made 9 changes total), until I got the file back to the same length as the original file, the last couple of changes I did by trial and error, until I could see the file was the same length as the original 7min.9secs.26667.

This post has been edited by db1989: Sep 24 2013, 18:46
Reason for edit: as in post #5
Go to the top of the page
+Quote Post
saratoga
post Sep 24 2013, 15:39
Post #8





Group: Members
Posts: 4996
Joined: 2-September 02
Member No.: 3264



QUOTE (bomber1978 @ Sep 24 2013, 08:44) *
Not sure if this is the correct explanation, but I made the speed changes by % eg. load in the file and make +3.0% change, then -2.5% change (made 9 changes total), until I got the file back to the same length as the original file, the last couple of changes I did by trial and error, until I could see the file was the same length as the original 7min.9secs.26667.


I think you misunderstood the question. Hes asking what method you used to change the speed, not how much you changed the speed by. I would begin answering this question by naming the program, and if possible, the algorithm used.
Go to the top of the page
+Quote Post
bomber1978
post Sep 24 2013, 15:58
Post #9





Group: Members
Posts: 27
Joined: 24-September 13
Member No.: 110284



QUOTE (saratoga @ Sep 25 2013, 00:39) *
QUOTE (bomber1978 @ Sep 24 2013, 08:44) *
Not sure if this is the correct explanation, but I made the speed changes by % eg. load in the file and make +3.0% change, then -2.5% change (made 9 changes total), until I got the file back to the same length as the original file, the last couple of changes I did by trial and error, until I could see the file was the same length as the original 7min.9secs.26667.


I think you misunderstood the question. Hes asking what method you used to change the speed, not how much you changed the speed by. I would begin answering this question by naming the program, and if possible, the algorithm used.


The particular files I posted I used Audacity 2.0.3.
I have no idea about algorithm..
Go to the top of the page
+Quote Post
bandpass
post Sep 24 2013, 16:50
Post #10





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



Audacity 2.0.3 uses the SoX algorithm (however, the change-speed dialogue box erroneously refers to libsamplerate). Assuming default settings, this means this graph from the src comparisons site applies:



The noise floor (faint blue) is that of 32-bit floats. Speeding up then down by a few % gives a slight hi-freq roll-off, but I doubt very much that it's audible; you could probably perform the conversion many 1000s of times and still not hear a difference.
Go to the top of the page
+Quote Post
bomber1978
post Sep 24 2013, 17:13
Post #11





Group: Members
Posts: 27
Joined: 24-September 13
Member No.: 110284



Hi,
I actually didn't use the default in Audacity 2.0.3, I turned "Shaped" dither to "None", as a did some mixdown tests and found that the only difference between Shaped dither and No dither was the dither was adding hiss (obviously), however there was no need for it when using 16bit files, there was NO difference in the mixdown apart from hiss...
(Bye the way in that process I actually discovered a bug Audacity acknowledged - using Shaped dither exporting as WAV add more hiss than exporting as FLAC!)
I presume using NO dither makes no difference to your graph and answer?

Also about the 2nd part of my question, if you don't think there will be an audible difference, what is creating the difference in the mixdown test I posted a sample of?, both original file and speed changed file are 7min9sec.26667, so why does the mixdown start off quiet and get louder as the time length progresses?
Thanks

This post has been edited by db1989: Sep 24 2013, 18:46
Reason for edit: ditto
Go to the top of the page
+Quote Post
Arnold B. Kruege...
post Sep 24 2013, 17:59
Post #12





Group: Members
Posts: 3812
Joined: 29-October 08
From: USA, 48236
Member No.: 61311



QUOTE (bomber1978 @ Sep 24 2013, 10:58) *
QUOTE (saratoga @ Sep 25 2013, 00:39) *
QUOTE (bomber1978 @ Sep 24 2013, 08:44) *
Not sure if this is the correct explanation, but I made the speed changes by % eg. load in the file and make +3.0% change, then -2.5% change (made 9 changes total), until I got the file back to the same length as the original file, the last couple of changes I did by trial and error, until I could see the file was the same length as the original 7min.9secs.26667.


I think you misunderstood the question. Hes asking what method you used to change the speed, not how much you changed the speed by. I would begin answering this question by naming the program, and if possible, the algorithm used.


The particular files I posted I used Audacity 2.0.3.
I have no idea about algorithm..



The answer to that question appears to be here"

http://wiki.audacityteam.org/wiki/SoundTouch

"
Audacity uses the SoundTouch library for the Change Pitch (changing pitch without changing tempo) and Change Tempo (changing tempo without changing pitch) effects. These are known as Time Stretching effects, because they defy the normal expectation that reducing the speed of the audio (thus increasing its length) will lower the pitch and vice-versa. Inevitably, some audible artifacts will be heard after applying these effects at more extreme settings.

Also, due to weaknesses in the algorithm, Change Pitch does not keep the length *exactly* as before. When increasing pitch, the processed audio tends to be reduced in
length very slightly. When reducing pitch, a little trailing silence is added at the end. This can make Change Pitch unsuitable for changing the pitch of individual very short notes. In a similar way, Change Tempo can remove a very short length of content from the start or end of the selection, or not stretch the end of the content to the end of the selection when slowing down. Both of these symptoms may leave a short silence at the end of the resultant selection.

Current Audacity now has an alternative time stretch effect called Sliding Time Scale/Pitch Shift. This uses an alternative time stretching library known as "Subband Sinusoidal Modeling" (sbsms). Sliding Time Scale / Pitch Shift allows you to change tempo (without changing pitch) while simultaneously changing pitch (without changing tempo). It can also set a different tempo or pitch change value for the start and end of the audio. Artifacts are still expected at more extreme settings, although no length changes arise from changing pitch, and changing tempo should not result in removed content or silences at the end of the selection. Note: this effect is several times slower than SoundTouch and might not work properly on low-powered machines.
"
Go to the top of the page
+Quote Post
bomber1978
post Sep 24 2013, 18:10
Post #13





Group: Members
Posts: 27
Joined: 24-September 13
Member No.: 110284



I think that is refering to "change pitch" and "change tempo", which are both differerent to the "change speed" function...
I have heard the Audacity experts/pros themselves say that their "change speed" function retains much better quality than their "change pitch" and "change tempo" functions...., but I guess the "speed change" function could also have similar slight differences in output file time length....

This post has been edited by db1989: Sep 24 2013, 18:42
Reason for edit: deleting pointless full quote of above post
Go to the top of the page
+Quote Post
bomber1978
post Sep 24 2013, 18:17
Post #14





Group: Members
Posts: 27
Joined: 24-September 13
Member No.: 110284



QUOTE (bandpass @ Sep 25 2013, 01:50) *
Audacity 2.0.3 uses the SoX algorithm (however, the change-speed dialogue box erroneously refers to libsamplerate). Assuming default settings, this means this graph from the src comparisons site applies:



The noise floor (faint blue) is that of 32-bit floats. Speeding up then down by a few % gives a slight hi-freq roll-off, but I doubt very much that it's audible; you could probably perform the conversion many 1000s of times and still not hear a difference.


So it uses SoX even though it says it uses libsamplerate?
Can you tell what both Audacity 1.2.6 and 2.0 used?, as I've used them also...
Thanks
Go to the top of the page
+Quote Post
uart
post Sep 24 2013, 19:29
Post #15





Group: Members
Posts: 810
Joined: 23-November 04
Member No.: 18295



QUOTE (bomber1978 @ Sep 24 2013, 10:17) *
So it uses SoX even though it says it uses libsamplerate?
Can you tell what both Audacity 1.2.6 and 2.0 used?, as I've used them also...
Thanks

Unfortunately earlier versions were significantly worse. Looking at the results HERE version 2.0 appears to be "broken".

This post has been edited by uart: Sep 24 2013, 19:29
Go to the top of the page
+Quote Post
bandpass
post Sep 24 2013, 20:21
Post #16





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



QUOTE (bomber1978 @ Sep 24 2013, 18:17) *
So it uses SoX even though it says it uses libsamplerate?

Yes, they used to use libsamplerate for some builds — there's some old info here: http://wiki.audacityteam.org/wiki/Libsamplerate
They switched to SoX in 2.0.3 — see here: http://wiki.audacityteam.org/wiki/libsoxr
but they neglected to update the text on the change speed screen (this text was removed in version 2.0.4).

Bottom line is that audacity 2.0.3 or above is absolutely fine for your needs.
Go to the top of the page
+Quote Post
Arnold B. Kruege...
post Sep 24 2013, 22:00
Post #17





Group: Members
Posts: 3812
Joined: 29-October 08
From: USA, 48236
Member No.: 61311



QUOTE (bomber1978 @ Sep 24 2013, 08:19) *
QUOTE (pdq @ Sep 24 2013, 21:48) *
I'm sure this is obvious, but if you can't ABX the result of nine time changes, doesn't that tell you that you are worrying about nothing?


Not necessarily, I know I have a fairly good ear for audio, but I can only do these tests on my laptop, which currently doesn't have the greatest soundcard, I know I can hear detail on my stereo system which I can't hear on my laptop with headphones, so it's not really ideal for me to do this test on the laptop I'm currently using...


You are not saying which audio interface that your laptop has, or what model your laptop is, so I guess we are supposed to be mystified.

I've had problems with the audio interfaces on some of my laptops having nastly noises and gurgles in their audio interfacts. When that happens I use a good external USB interface with a good headphone amp.

The actual detail you hear in different audio interfaces that are free of obvious faults is vastly overstated by people who have zero experience with good subjective testing.

No headphone comes within an order of magnitude of having as low distortion as a good audio interface, and in general ears aren't as good as headphones. Notice that they test ears with headphones and not vice-versa!

QUOTE
And I know there are people here who know what digital artefacts introduced by speed changing may sound like, whereas I really don't know what to listen for..., does it sound the same as the digital "underwater / metallic" noise that noise/hiss reduction introduces?, or does it sound like something else?, for example a friend thought they could hear some echo/reverb on the speed changed file?, but I can not hear that?, so maybe there's something there to be heard which I can not hear?, once something is pointed out to me I can usually hear it and learn about it, I'm also using this as a way to try and learn more, so if there's something there which is audible I'd like to know!
Plus like I said, I do hear a difference in the mixdown, however I don't know if that is because of drift, or if it's because of difference in quality? - which was part of my question?, and reason I posted asking for help & opinion...
So I'm really hoping someone can give an opinion who has the proper equipment / ear to give an answer...


Shouldn't be any audible artifacts.

The fact that your signal subtraction test came out with a fairly giant residual doesn't prove anything. The problem with subtraction tests is that they magnify small changes, even changes that you are never going to hear. So they are good for unecessarily raising a lot of concerns and IMO not a heck of lot else.

Do some blind tests with the ABX plug in for Foobar, and move on if they come out well.
Go to the top of the page
+Quote Post
bomber1978
post Sep 25 2013, 06:06
Post #18





Group: Members
Posts: 27
Joined: 24-September 13
Member No.: 110284



QUOTE (uart @ Sep 25 2013, 04:29) *
QUOTE (bomber1978 @ Sep 24 2013, 10:17) *
So it uses SoX even though it says it uses libsamplerate?
Can you tell what both Audacity 1.2.6 and 2.0 used?, as I've used them also...
Thanks

Unfortunately earlier versions were significantly worse. Looking at the results HERE version 2.0 appears to be "broken".


Hi,
I can't tell what that graph for Audacity 2.0 exactly means (the one you say appears to be "broken").
I have been using it (2.0) for the past year, are the results you have seen relating to 2.0 something which would be audible?
Thanks
Go to the top of the page
+Quote Post
bandpass
post Sep 25 2013, 07:15
Post #19





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



It's all about likelihood: the more distortion introduced, the more likely it is to be audible—dependent on the source material (simple sine waves at one extreme; Metallica at the other) and what happens to the audio subsequently (such as adjusting tone controls at playback).

BTW, the two resamplers previously used by Audacity were not tied to specific versions, but to specific builds (typically, Linux distros built with one, Mac & Windows built with the other). In all cases, old and new, the quality also depends of the preference settings.

AFAIK, the 'broken' Audacity 2.0 resampler actually works to spec, it just hails from a day when CPU cycles were few and far between—it's a bit below CD spec, but not horribly so.
Go to the top of the page
+Quote Post
Juha
post Sep 25 2013, 07:28
Post #20





Group: Members
Posts: 460
Joined: 14-February 07
From: EU-FIN
Member No.: 40610



Have you tried alternative software for your SRC tests? Still got the same result in mixdown?
Voxengo r8brain is one you could try - http://www.voxengo.com/product/r8brain/

Juha
Go to the top of the page
+Quote Post
bandpass
post Sep 25 2013, 08:08
Post #21





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



QUOTE (bomber1978 @ Sep 24 2013, 17:13) *
I presume using NO dither makes no difference to your graph and answer?

Dither is a separate subject; resampling/speed-change tests/graphs are generally done at bit-depths that don't require/use dither (as were mine and the ones linked to).
QUOTE (bomber1978 @ Sep 24 2013, 17:13) *
Also about the 2nd part of my question, if you don't think there will be an audible difference, what is creating the difference in the mixdown test I posted a sample of?, both original file and speed changed file are 7min9sec.26667, so why does the mixdown start off quiet and get louder as the time length progresses?
Thanks

Because the speed-up and speed down are not exactly matched: the slightest difference will very quickly result in a huge difference in a 'null' test.

Your best chance is to use reciprocals: if you change speed up by 3%, then change it by 100*(1/1.03-1) = -2.9126213592233% to get it back (may still be subject to some rounding error, fixed group delay, and there will be some HF roll-off, reducible by changing the order to speed-down then speed up).
Go to the top of the page
+Quote Post
uart
post Sep 25 2013, 13:12
Post #22





Group: Members
Posts: 810
Joined: 23-November 04
Member No.: 18295



QUOTE (bomber1978 @ Sep 24 2013, 22:06) *
QUOTE (uart @ Sep 25 2013, 04:29) *
QUOTE (bomber1978 @ Sep 24 2013, 10:17) *
So it uses SoX even though it says it uses libsamplerate?
Can you tell what both Audacity 1.2.6 and 2.0 used?, as I've used them also...
Thanks

Unfortunately earlier versions were significantly worse. Looking at the results HERE version 2.0 appears to be "broken".


Hi,
I can't tell what that graph for Audacity 2.0 exactly means (the one you say appears to be "broken").

Go to that link and select "Audacity 2.0 (High Quality)" in the converter drop down and compare (leave the test set to sweep). All the purple and red stuff that you can see there (other than the main white trace) is essentially distortion - as in harmonic components that shouldn't be there.

QUOTE
I have been using it (2.0) for the past year, are the results you have seen relating to 2.0 something which would be audible?

By "broken" I only meant that it is significantly worse than other contemporary sample rate converters. As bandpass points out, it's still probably inaudible at about -80 dB. The best resamplers however have those harmonic levels down around -140 to -150 dB level (pretty much consistent with 24 bit digital silence). With levels like that you don't even attempt any kind of listening test, because you know it's futile.

This post has been edited by uart: Sep 25 2013, 13:14
Go to the top of the page
+Quote Post
bomber1978
post Sep 25 2013, 13:31
Post #23





Group: Members
Posts: 27
Joined: 24-September 13
Member No.: 110284



Agh I think I just wasted my time, when you said it was "broken" I thought you meant there would be audible quality loss, so I changed a file 9 times with 1.2.6 and loaded it onto my desktop with 2 of the original files and tried 20 times to pick the odd one out (the speed changed file), I got it correct 7/20 times so I guess nothing definitive there.... However results were completely different when I set quality from "HQ" to "fast sync", I got it right 10/10.

This post has been edited by db1989: Sep 25 2013, 18:57
Reason for edit: deleting pointless full quote of above post
Go to the top of the page
+Quote Post

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: 30th September 2014 - 20:03