Current Version Problem?

2005-02-17 15:46:33

OK i covered this before but it wasnt really solved/answered before.  I discovered a problem sample for vorbis - can be found here:    (I didnt go looking for this sample i found it on the 2nd CD i was encoding at the time)

http://www.rarewares.org/quantumknot/Originalsmall.flac

Topics

http://www.hydrogenaudio.org/forums/index....showtopic=23424

http://www.hydrogenaudio.org/forums/index....92&#entry225892

Sounds awful, even at Q5 and still abx'able at q7.    Some of QK options for tuning the impulse trigger profile and impulse noise_tune help if adopted - but at big bitrate penalties and dont make a huge amount of difference.    OK all of this is fair enough, there will always be problem samples. 

However as I previously mentioned the old GT2 tuning of the RC2 encoder doesnt have this problem at all even at bitrates as low as 128!  This is quite a discrepancy and a huge step backwards for my money.    I guess im asking what could be different in the old RC2 based encoder that doesnt create the sample problem as with any of the newer based ones?      Sorry to bring this up again im really not trying to be a nag!    Perhaps no one knows.  I just am genuinely astounded at such a retrograde step.   

I believe QK is planning some new things when he has time again - are these things that would help in this area QK?

Current Version Problem?

Reply #1 – 2005-02-17 23:45:44

Yes, the smearing of microattacks is something I intend to focus on alleviating.  I'm not familiar with the internals of RC2 or GT2 so I can't comment on why the problem is not there (perhaps Garf remembers a bit of his previous work  ), but just to clarify, when using RC2 and GT2, is the problem of smearing not there or is it just very very small?  Also, does the bitrate jump around a lot?

I wonder how much difference is there between the block switcher of RC2 and the final one.

Current Version Problem?

Reply #2 – 2005-02-18 07:07:39

The RC2 encoder is quite different from 1.0 in high bitrate behaviour (and as noted before, it can be better than 1.0 quite a few times), so I'm afraid the matter is going to be more complex than 'figure out the tuning to copy'.

Current Version Problem?

Reply #3 – 2005-02-18 09:46:38

QK - the bitrate (i think i menioned before) actually jumps around less with GT2.  I previously tried to use the managed mode to curb the jumping around bitrate, which really did seem to help the problem a lot.

The problem isnt there with any of the three bitrate settings for GT2 (128, 160, 350 i believe).  (or if it is at 128 its vanishingly small)

And even at the 128 setting the result is preferable to even q6 with impulse set to -5 and trigger profile at 3 (just a quick test as i havent got time for more) with huge resulting increase in bitrate (although it does help some)

--managed -b 165 -m 155 -M 200 --advanced-encode-option impulse_noisetune=-5 --advanced-encode-option impulse_trigger_profile=3

That command line used with the latest oggenc 2.3 produces much better results on the problem (bizarrely).

Not sure if thats any help.  Hope you (and others too) can continue making good progress when time allows.

EDIT:

tried this: --managed -b 130 -m 120 -M 150 --advanced-encode-option impulse_noisetune=-1 --advanced-encode-option impulse_trigger_profile=4

Gives a bitrate very close to GT2 128 setting and sounds fine with no trace of the problem.