As I have mentioned in my other thread, I dont recommend using ITP for general encoder anymore since I've heard reports of it degrading quality in some instances.  Hence use ITP as a problem fixer (noise in microattacks), rather than a general quality improver.  If in doubt, don't use it at all.

INT seems safe for now
[a href="index.php?act=findpost&pid=245161"][{POST_SNAPBACK}][/a]

Can you please be more specific?
In which cases it may cause problems? I'm using q4 with ITP set for 3, is that OK?
[{POST_SNAPBACK}][/a]

Quote

Can you please be more specific?
In which cases it may cause problems? I'm using q4 with ITP set for 3, is that OK?
[{POST_SNAPBACK}][/a]

Unfortunately, I can't be anymore specific since I haven't encountered and verified this yet.  I got the report from [a href="http://nyaochi.cocolog-nifty.com/audio/2004/09/ogg_vorbis_11_.html]here[/url].  Maybe the best person to explain the problem is nyaochi

I need some further verification on this.  Hence, to be on the safe side, ITP is no longer recommended as a quality improver, but more of a problem fixer.  So only use it if you hear microattack noise.  There have been plenty of listening tests that have verified ITP to be helpful in microattack samples so I'm confident in that respect.
[a href="index.php?act=findpost&pid=245169"][{POST_SNAPBACK}][/a]

OK, thank you for the quick reply.
Personally I still haven't encountered this problem myself.. but I'm starting to feel a bit uncomfortable. The thing is that my music collection is huge, so I gotta make sure that I'm ripping it right.

During those listening tests that have verified ITP to be helpful in microattack samples, were there any reports on problems that ITP is causing?

Sorry if I'm a hassle (being a newbie and all..), but this is really important for me.

Quote

OK, thank you for the quick reply.
Personally I still haven't encountered this problem myself.. but I'm starting to feel a bit uncomfortable. The thing is that my music collection is huge, so I gotta make sure that I'm ripping it right.

If in doubt, don't use ITP at all.  It only addresses one problem and depending on the genre, it may be quite rare.  So out of your music collection, 98% may sound fine without ITP but there may be the odd couple which exhibit some noise.  Hence you could re-encode those few songs with a higher ITP setting.  Hence, its better to be safe on the majority and re-encode the few, if need be. 

INT, however, seems rather safe to use and will generally reduce pre-echo.  In fact, what INT does is exactly how I tune pre-echo in the QKTune, though tweaking individual values is more flexible than offsetting which is what INT does.

Quote

During those listening tests that have verified ITP to be helpful in microattack samples, were there any reports on problems that ITP is causing?

I myself did a successful ABX test with this sample that Rotellian provided

http://rarewares.soniccompression.com/quan...ginalsmall.flac

I mainly focused on the noise generated by the microattacks but didn't pay attention to other artifacts.  It is quite a good sample for getting familiar with noise due to microattacks so encode it normally at q 4 and then do an ABX on it.  Try to listen for that puff of noise.  Then increae ITP and it should reduce a bit.

Quote

Sorry if I'm a hassle (being a newbie and all..), but this is really important for me.

Feel free to raise anything and give feedback.  It is mainly because of the lack of feedback on ITP that led to my mistake of using it as a recommended option.  Now that listening tests have come to light, I now know that isn't good to use on everything.
[a href="index.php?act=findpost&pid=245171"][{POST_SNAPBACK}][/a]

I did ABX on the sample you gave me:
It was awfully easy, even at q6...
However, ITP set to 3 solved the problem completely!
BTW, has nyaochi provided samples and settings information?
Not that I'm questioning his credibility, I just couldn't find it on his site.
[a href="index.php?act=findpost&pid=245176"][{POST_SNAPBACK}][/a]

ABC/HR Version 1.0, 6 May 2004
Testname: 8823.flac

1L = 8823.flac-1.1.wav
2R = 8823.flac-1.1-itp3.wav
3L = 8823.flac-1.1-aotuv.wav

---------------------------------------
General Comments:

---------------------------------------
2R File: 8823.flac-1.1-itp3.wav
2R Rating: 3.0
2R Comment: Noisy (distorted reverberation sound of hi-hat)
---------------------------------------
3L File: 8823.flac-1.1-aotuv.wav
3L Rating: 4.0
3L Comment: hi-hat
---------------------------------------
ABX Results:
Original vs 8823.flac-1.1.wav
    3 out of 7, pval = 0.773
Original vs 8823.flac-1.1-itp3.wav
    7 out of 8, pval = 0.035
Original vs 8823.flac-1.1-aotuv.wav
    22 out of 33, pval = 0.040

Quote

I did ABX on the sample you gave me:
It was awfully easy, even at q6...
However, ITP set to 3 solved the problem completely!
BTW, has nyaochi provided samples and settings information?
Not that I'm questioning his credibility, I just couldn't find it on his site.
[a href="index.php?act=findpost&pid=245176"][{POST_SNAPBACK}][/a]

I also confirmed that ITP significantly improved this kind of samples (e.g., badvilbel.flac and Originalsmall.flac). I could easily ABX it as you did. The effect of ITP is fascinating for them. But these samples may be extreme. I've never found this problem in my music collection.

We should listen to more samples to make a conclusion. That's why I tried to ABX in my private diary. However, I gave up comparing because it was very hard to rate these samples (i.e., samples with/without ITP). They were so close that I had to spend much time for training and listening. I didn't have so much time at that time and gave up  . In my limited numbers of listening samples, I felt strange behavior of ITP. In 8823 sample, for instance,

Code: [Select]

ABC/HR Version 1.0, 6 May 2004
Testname: 8823.flac

1L = 8823.flac-1.1.wav
2R = 8823.flac-1.1-itp3.wav
3L = 8823.flac-1.1-aotuv.wav

---------------------------------------
General Comments:

---------------------------------------
2R File: 8823.flac-1.1-itp3.wav
2R Rating: 3.0
2R Comment: Noisy (distorted reverberation sound of hi-hat)
---------------------------------------
3L File: 8823.flac-1.1-aotuv.wav
3L Rating: 4.0
3L Comment: hi-hat
---------------------------------------
ABX Results:
Original vs 8823.flac-1.1.wav
    3 out of 7, pval = 0.773
Original vs 8823.flac-1.1-itp3.wav
    7 out of 8, pval = 0.035
Original vs 8823.flac-1.1-aotuv.wav
    22 out of 33, pval = 0.040

So I'm not for or against ITP. If you encounter this problem as in Originalsmall.flac frequently in your music collection, the importance of ITP will be high. And that's why I didn't post this on HA. 

EDIT: grammar
[a href="index.php?act=findpost&pid=245207"][{POST_SNAPBACK}][/a]

The ITP switch has been out in the public for a month or so and most of the listening tests reported positive results, in terms of reducing noise in these microattack samples.  Also, GT3b2 makes use of slightly different profiles as well, hence the impression at the time was that ITP is a harmless switch to use.  It also filled in the gap (along with INT) left when retiring GT3b2 and allowed us to use a single encoder recommendation.  Hence I don't see how I could've done things differently, given the situation.   I already delayed updating the thread for quite a few months now, up until 1.1 was released.

The reason I reacted in such a way when reading nyaochi's test result is that it was the first and only report of ITP causing a regression and that was enough reason for me to attach a degree of caution to using it, esp. when it is in a thread about recommended settings.  Before this report of regression, it seemed ok to recommend it since it was a win-win situation anyway ie. you dont have the noise problem in the first place but it doesn't hurt to use it anyway.  But if it causes regression, even in one sample, then we need to be more cautious about recommending for general use.  Since ITP is only useful at fixing a specific problem but adds rather little in general, perhaps it is better to not include it in the recommendations thread.
[a href="index.php?act=findpost&pid=245236"][{POST_SNAPBACK}][/a]

I should add that nyaochi's result has quite a bit of weight in my opinion, since he is another third-party Vorbis developer.
[a href="index.php?act=findpost&pid=245242"][{POST_SNAPBACK}][/a]

I should add that nyaochi's result has quite a bit of weight in my opinion, since he is another third-party Vorbis developer.
[a href="index.php?act=findpost&pid=245242"][{POST_SNAPBACK}][/a]

(...) aoTuV beta 4 introduces almost the same result as if you'd tinker the 1.1 ITP's to death in search for the proper quality/size and microattacks/noise values.

ABC/HR Version 1.0, 6 May 2004
Testname: 

1L = C:\temp2\preecho\castanets2 [1.1.1 q6 IN-15].wav
2R = C:\temp2\preecho\castanets2 [aoTuVb4 q6].wav
3R = C:\temp2\preecho\castanets2 [aoTuVb4 q7].wav

---------------------------------------
General Comments:

---------------------------------------
1L File: C:\temp2\preecho\castanets2 [1.1.1 q6 IN-15].wav
1L Rating: 4.8
1L Comment: 
---------------------------------------
2R File: C:\temp2\preecho\castanets2 [aoTuVb4 q6].wav
2R Rating: 4.0
2R Comment: 
---------------------------------------
3R File: C:\temp2\preecho\castanets2 [aoTuVb4 q7].wav
3R Rating: 4.5
3R Comment: 
---------------------------------------
ABX Results:
C:\temp2\preecho\castanets2 [1.1.1 q6 IN-15].wav vs C:\temp2\preecho\castanets2 [aoTuVb4 q6].wav
    5 out of 5, pval = 0.031
C:\temp2\preecho\castanets2 [1.1.1 q6 IN-15].wav vs C:\temp2\preecho\castanets2 [aoTuVb4 q7].wav
    5 out of 5, pval = 0.031
C:\temp2\preecho\castanets2 [aoTuVb4 q6].wav vs C:\temp2\preecho\castanets2 [aoTuVb4 q7].wav
    5 out of 5, pval = 0.031

1.1.1 -q6 -q --advanced-encode-option impulse_noisetune=-15