Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: New experimental version of aoTuV (Read 14896 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

New experimental version of aoTuV

Reply #25 –
Quote
I think that this problem is in block switching.
For example, this of the noise of brass is just the cause.
[a href="index.php?act=findpost&pid=239870"][{POST_SNAPBACK}][/a]

Some brass instruments require short-block (it's obvious with a sample like Orion II - trombone instruments). Noise appear for that reason (and disapear with GT3b2 for exemple, which add more bits on critical parts).

But there's also audible noise on tonal instruments (violins, cellos,...), and I don't think that block switching could be involved. But I'm maybe wrong - it's just a feeling.
If I find some time, I'll compare recent aoTuV with fully uncoupled release of vorbis on samples which exhibits this problem.
Wavpack Hybrid: one encoder for all scenarios
WavPack -c4.5hx6 (44100Hz & 48000Hz) ≈ 390 kbps + correction file
WavPack -c4hx6 (96000Hz) ≈ 768 kbps + correction file
WavPack -h (SACD & DSD) ≈ 2400 kbps at 2.8224 MHz

New experimental version of aoTuV

Reply #26 –
Quote
It seems that aoTuV (and perhaps other third party tunings) will become the main focal point of Vorbis tuning improvements with Monty now taking on the management role of Executive Director and no-one at Xiph.Org to do any coding.
[a href="index.php?act=findpost&pid=239746"][{POST_SNAPBACK}][/a]


Argh... where did you hear that Monty is dropping out of coding?  I do so hope that's not true!

Aoyumu has done a nice job with tuning (and it's great to see that he [or she] is still tuning) as have the other tuners, but now that really leaves the ship without a captain.  Vorbis development was somewhat moribund before these tunings started popping up (I imagine the tunings have appeared largely out of frustration with the Xiph rate of progress), but for now all of the tunings do seem to be 'merely' tunings and there have been no clear post-1.0 leaps in the technology (compared with even LAME).  Monty seemed to have had Big Ideas at some point, about future improvements to the encoder, but they've only been very vaguely sketched out...

(On the subject of moribund Xiph projects, look at poor Theora... I think it's been in a pre-releases cycle for years!  It's a credit to the maintainers and the original VP3 donated code that the technology still looks even vaguely competitive after all this time.)

New experimental version of aoTuV

Reply #27 –
Quote
But there's also audible noise on tonal instruments (violins, cellos,...), and I don't think that block switching could be involved. But I'm maybe wrong - it's just a feeling.

If the noise is the kind which appears very momentarily, the possibility of problems other than block switching is large.
Otherwise, I suspect block switching. The problem of point stereo will be raised as other possibilities.

Quote
If I find some time, I'll compare recent aoTuV with fully uncoupled release of vorbis on samples which exhibits this problem.

Though regrettable, aoTuV exp has not improved dramatically with this kind of sample. The block switching algorithm itself is because it is not improving. It is a future subject.


Quote
where did you hear that Monty is dropping out of coding? I do so hope that's not true!

Monty did not necessarily stop development of Vorbis. He will be going to make new psymodel in the future (it heard from him). However, he is a very busy man....

New experimental version of aoTuV

Reply #28 –
Quote
Quote
You can easily hear this noise issue by generating a tone in Audicity and then encoding the resulting WAV with Vorbis.

Sine wave?
Would you upload a sample, if very well?
[a href="index.php?act=findpost&pid=239870"][{POST_SNAPBACK}][/a]

A sine wave...but I did more experimentation and found that the noise problem exists only with sine tones with amplitudes very close to 1 (>0.98). The noise goes away if you create a tone with amplitude of 0.95 or less (didn't try all frequencies however). So the problem is likely due to clipping from quantization. I know LAME normalizes some inputs to 95-98% to deal with this issue...though that's not necessarily the solution here.

This noise problem is not local to aoTuV but Vorbis in general.

New experimental version of aoTuV

Reply #29 –
if i remember right, then the noise problems appeared first in 1.0 final..... and where not there in 1.0 RC3.

I dont know how many thing got changed between those two versions - i guess lots of them. Yet still, couldn't this just be used for an "find the culprit by systematic exclusion"-approach? Thus, reverting parts of the code to old RC3, one after another, until the culprit is found?

- Lyx
I am arrogant and I can afford it because I deliver.

New experimental version of aoTuV

Reply #30 –
Quote
Argh... where did you hear that Monty is dropping out of coding?  I do so hope that's not true!


I got it from the minutes of the xiph meeting

http://westfish.xiph.org/~giles/200408_meeting.txt

Quote
18:08 < Garf> who's going to improve vorbis then?
18:08 < xiphmont> People take over coding maintainence or nothing happens.  But I do think there's enough work being submitted now to keep it moving.
18:08 < xiphmont> Garf: for the time-being, whoever steps up and not me.  I'll continue docs, but no new code for a while.


However, things look slightly promising from the latest minutes

http://westfish.xiph.org/~giles/200409_meeting.txt

Quote
17:05 < Arc> does this mean monty is no longer primarily a coder?
17:06 < rillian> I'm also now 'treasurer' so in the unlikely event someone has financial questions, you can direct them to me
17:06 < xiphmont> That means I will function as an executive for the organization, not as technical staff.
17:06 < xiphmont> Arc: correct.
17:06 < Arc> so the bugfixes you made to libogg recently were your last for awile?
17:06 < xiphmont> No, it's simply not my primary duty.


I guess Monty hasn't totally moved away from coding but moved it to a 'secondary duty', but typically when you take on a management position, it occupies nearly all of your time.  When Monty was the primary coder, development was already quite slow, but now he has more stuff on his plate, I think it is going to be even slower.

New experimental version of aoTuV

Reply #31 –
Ugh.  What is Monty thinking?  Vorbis development is already too slow.  Unless he's willing to focus and do one thing and do it well all of the Xiph projects are going to wind up as interesting historical footnotes.  Score another one for aac.

New experimental version of aoTuV

Reply #32 –
lets face it, the only quality improvements for 1.1rc1 came from Aoyumi.

I was excited the time Monty announced something like the ultimate HE-AAC killer code, but this won't happen anytime soon now...

This doesn't mean I don't care or fail to see what Monty created, to the contrary!
 

New experimental version of aoTuV

Reply #33 –
Quote
A sine wave...but I did more experimentation and found that the noise problem exists only with sine tones with amplitudes very close to 1 (>0.98). The noise goes away if you create a tone with amplitude of 0.95 or less (didn't try all frequencies however). So the problem is likely due to clipping from quantization. I know LAME normalizes some inputs to 95-98% to deal with this issue...though that's not necessarily the solution here.

This noise problem is not local to aoTuV but Vorbis in general.
[a href="index.php?act=findpost&pid=240100"][{POST_SNAPBACK}][/a]

I understand it.
Probably, use of Vorbis Gain (replay gain) will be appropriate in order to solve this problem completely.