Artifact test

[Aty test]

I tested Audacity versions on 2 different PC's Win XP and one fresh installed Win 7 machine

When change Pitch from 440.000 to 432.000 the next Versions show artifacts in the ouput.
You clearly can hear an instability in the output file.
You need a good wav example and speaker system to hear the difference.

1.23
1.24
1.24b
1.26
1.3.13 beta

Version 1.22 has no encoding distortion i could find.
Works both on Win Xp and Win 7.

Go here:

http://www.oldversion.com/Audacity.html


Optional for testers:

Audacity artifact test.wav file.

http://www.megaupload.com/?d=S7K170YW

[steve]

It is advisable to always download programs from the official sources.
Old versions are not recommended for general use, but are available for testing purposes from here: http://sourceforge.net/projects/audacity/files/

There's only one sound in the linked file. Do you have a "before and after" comparison of the issue that you are trying to demonstrate?
Also, could you describe in detail your test procedure.

(I've just tried using "Change Pitch" on a sine wave, to change from 440 Hz to 432 Hz and it's virtually perfect with Audacity 1.3.14 alpha).

[Aty test]

This wav file is after encoding from 440.000 to 432.000 hz it is distorted.

http://www.megaupload.com/?d=1NKVBCKK

If you try a sine wave of 300 hz generated in Audacity there is no distortion after encoding i can confirm,
i already tried that when testing.
The wav file in the first post is part of the beginnig of a music track, with lots of harmonics and more complex that's the difference.

[steve]

"Time Stretch" effects (change pitch / change tempo) are work by chopping up the sound into little pieces, stretching them out or pushing them closer together, then sticking all of the pieces back together again. Exactly how best to do that is a matter of optimisation and compromise.

I agree that on that particular sample, the change pitch effect in Audacity 1.2.2 works better than later versions, but I've also tested a number of other samples where the later version works better than the 2.2 version. There is never going to be a perfect compromise that suits all source material.

In order to achieve sound stretching, Audacity uses a third party "library" called SoundTouch http://www.surina.net/soundtouch/
I would guess that the difference between Audacity 1.2.2 and later versions of Audacity are due to using a later version of SoundTouch.
However, Audacity 1.3.13 now has an additional method of time stretching that uses an entirely different method of stretching that is usually better than either of these earlier version (but slower). See here for details: http://wiki.audacityteam.org/index.php?title=SoundTouch

[Aty test]

I still have my doubts, on most music tracks that play with lost of instruments modern music etc.
than it is very difficult to notice any distortion that results from encoding from 440 to 432 hz but there is.
You have to find a music track that sounds silent with minor instruments than you could sense a distortion,
after that do the same with version 1.22 to check if the distortion is still there or not.

If i knew a way to compare wav files with software i might do so, but i don't know a reliable way.
You could pitch it back from 432 to 440 and compare it with the original, that is only an idea.

[steve]

It's definitely interesting that there is such a noticeable difference with that audio sample, and I'm quite surprised that it sounds so bad on recent versions of Audacity, but as you say, you need to choose the right (wrong) kind of audio samples to show up the issue.

Taking another example, with a pure sine wave, you agree that the stretching is extremely good with recent versions of Audacity, but with 1.2.2 there is a very noticeable deterioration of sound quality.

For a small pitch shift such as 440 > 432 it may be better to use the Change Speed effect. The sound quality is very much better than any pitch shifting algorithm, though of course the tempo will also change a little, but if that is acceptable then it would produce the best results.

[Aty test]

Audacity uses the Soundtouch library there are 2 algorythms,
Time-stretching (heavier operation pitch stays the same) and Sample Rate Transposing ( stretching
and affected pitch)

Now i have tested versions of Audacity the time needed for a single wav file to encode

- Changing Pitch from 440 to 432 which is a combination those (time-stretching and sample rate transposing)

v.1.20 109 sec.
v.1.22 109 sec.
v 1.23 19 sec.
v 1.24b 22 sec.

v.1.3.13 15 sec. (Subband Sinusoidal Modeling in stead of Soundtouch)


- Time stretching test Pitch not affected, 1.0 % longer

v 1.22 95 sec.
v 1.23 23 sec.
v 1.24b 19 sec.

v 1.3.13 20 sec. (Subband Sinusoidal Modeling in stead of Soundtouch)

Comment:
Interesting to see is the old 1.2 versions take their time for encoding, from 1.23 suddenly its a
lot faster (maybe not as accurate).

v 1.3.13 uses an alternative time stretching library known as "Subband Sinusoidal Modeling"
it says it is several times slower than SoundTouch, but i don't see how with the testresults.

[steve]

v 1.3.13 uses an alternative time stretching library known as "Subband Sinusoidal Modeling"
it says it is several times slower than SoundTouch, but i don't see how with the testresults.

Subband Sinusoidal Modeling is used by the effect "Sliding Time Scale/Pitch Shift". This is a new (additional) effect in recent versions of Audacity 1.3.