Any Quality difference between OggMachine, HeadAC3he, oggenc?

[kagoru]

Does someone know if there's a quality difference when using oggENC with a wave created with azid compared to the conventional methods like HeadAC3he or OggMachine?

TIA.

[MaTTeR]

I doubt you would hear a difference but I haven't personally did a listening test. If you do, please post your results.

[canadian_fbi]

note: this is more of a potential bug report than a direct reply to the first post, but it's on the subject more or less so what the hey.

this is more of a file size/bitrate issue than a quality issue (though it should affect quality too in theory), but i've done two movies with headac3he at quality 4.99 that have both come out way larger than they should be (203 kbps and 188 kbps). the 203 kbps file i was content to reencode at a lower quality level, thinking that there may have been something in the movie that led to the higher bitrate, but when the second movie came out at 188, i downloaded the oggmachine and tried it at 4.99 quality there too - it ended up at a normal 117 kbps bitrate.

this is using headac3he 0.22b with the new vorbis.dll posted on 3/20 on darkavenger's site. i suspect it's the newly optimized dll causing the problem, though i haven't ran any tests to see if restoring the old one fixes the problem. will do that now. besweet under the almighty control of the oggmachine does it as expected though.

[MaTTeR]

canadian_fbi,

What version is your DLL? The log window should display it when you launch HeadAC3he. I ask because I've encoded a few flicks in the last few days using the faster DLL without the bitrate problem.

[canadian_fbi]

the old dll (vorbis.dll 1.11 (06.03.2002)) gave an output of about 117 kbps, a size difference of only 20 KB over the oggmachine version.

the new dll is listed as vorbis.dll 1.11 (18.03.2002).

[MaTTeR]

Here is mine-
vorbis.dll 1.11 (22.03.2002) engine 1.66(This is alpha though)

EDIT- There were some issues in the 03-18 build from Roberto but I thought it had something to do with compiling, perhaps it's related. I think D A will be releasing the 03-22 build tomorrow providing no probs crop up tonight.

PM me your email addy

[DarkAvenger]

It probably is not a bug, as a lot of psycho accoustics have changed with latest dll, thus enhancing speed. Ty getting latest dll for OGGMachine, as well. It should give you similar results. There is not much I can do wrong when I compile the dll, I guess... Could you check whether you get different filesize, when you sellect ligh lossy and lossless channel coupling? That would be the only thing where I could do crap.

[canadian_fbi]

with stereo coupling set to lossless the bitrate went up to 214. this is at -q 4.99 mind you. if this is a psychoacoustics issue i don't think i like the change :)

[canadian_fbi]

here are some results of some tests i ran with the various dlls:
(the test subject was the ac3 from the kung fu chapter of the matrix, length 4:33, encoded at 4.99 quality)


dll version dumb time float time file size
----------- --------- ---------- ---------
old (3/06) 1:07 1:03 4107 KB
VC6 (3/18) 1:07 1:03 6677 KB
ICL (3/22) :50 :46 3882 KB

OggMachine 1:09 N/A 4106 KB
notes:
the difference in the file sizes between dumb and float were negligible, so the dumb file size is listed.
3/06 is the old dll i had prior to downloading the 3/18 version.
the 3/18 dll is the same as the MSVC6 compile listed on dark avenger's dl page as of now under the date 23.03.2002 (from what i can tell, anyway).
3/22 is what's currently up as the ICL compile under the date 22.03.2002.
when i tried substituting the dll for oggmachine, it wouldn't work. i was switching vorbis.dll in the windows/system32 directory - if this isn't the proper one to switch, let me know, because i couldn't think of any place better to try.

conclusions:
- the VC6 version is terribly screwed up, at least for me anyway.
- the ICL version is 25% faster and gives a 5% smaller file size.
- oggmachine gives basically the same output as the old dll, with a slightly longer processing time.

i haven't done any listening tests to the finished vorbis files yet, but i'm guessing the differences between the ICL and old dlls will likely be too small for me to reliably notice. i guess that new ICL compile's going to have itself a job :)

[MaTTeR]

Very impressive results. Thanks for the feedback.

when i tried substituting the dll for oggmachine, it wouldn't work. i was switching vorbis.dll in the windows/system32 directory
I think you need to add the libvorbis.dll into the BeSweet folder.

Did I mention the speed boost was impressive?:D

[DSPguru]

Originally posted by canadian_fbi
when i tried substituting the dll for oggmachine, it wouldn't work. i was switching vorbis.dll in the windows/system32 directory - if this isn't the proper one to switch, let me know, because i couldn't think of any place better to try.

- oggmachine gives basically the same output as the old dll, with a slightly longer processing time.if you're into comparing DLLs. download vorbis dlls (http://www.inf.ufpr.br/~rja00/files/OVdllswinlame.zip) and extract them to your windows/system directory.
then you can compare.

[canadian_fbi]

with those dlls that dspguru linked to, oggmachine's time was :59.

i thought that oggmachine used the oggds filters for encoding, and not any dlls in the besweet directory (because there don't seem to be any in there now and it encodes fine). in which case it seems that the encoding speed as far as the dlls are concerned would be directly related to the speed of the dlls in tobias's pack. in any case, simply unzipping those dlls into my windows/system directory did provide a good speed boost to oggmachine, but not quite to the level of the ICL compile.

dspguru: am i right about the dlls used in oggmachine? the ones you linked to have the latest optimizations, i'd assume, though i'm not sure how they were compiled.

[MaTTeR]

@DSPguru

All 4 of the dll files need to be in the sys32 dir? I was thinking only libvorbis:confused:

[DSPguru]

to make things clearer :
BeSweet (OggMachine) is using ogg.dll,vorbis.dll,vorbisenc.dll .
OggDS uses the same dlls, and therefore i don't supply them - i just ask ppl to make sure that OggDS is already installed.
Tobias told me that he recently made a fast compiles of the above dlls (he was talking about 100% speed improvements :confused: !!), so whenever he release them, you'll see the improvements also with BeSweet.

the above dlls are taken from the Vorbis RC3 SDK dlls. and it hold all functions needed for encoding Ogg Vorbis.
there's also a release called libVorbis.dll which is a single dll that merge the whole library into one.

HeadAC3e's vorbis.dll is the CDEX vorbis encoder which also based on libvorbis, but is different from the SDK dlls.
therefore you cannot switch between them.

EDIT:
i forgot to mention that the link i posted was to the latest compile of vorbis (meaning, you'll get the most updated quality implementations), but they're not ICL compiles (meaning, they're not so fast).

[MaTTeR]

Wheew...I'm sick of dll's already. LOL

Just kidding. Thanks for the detailed response though. That certainly clears things up for me.

Tobias told me that he recently made a fast compiles of the above dlls (he was talking about 100% speed improvements !!),

Now I'm drooling.

[canadian_fbi]

thanks dspguru, that does make things a lot clearer.

just a note about switching those dlls - using the newer ones did speed up oggmachine encoding, but when playing back an ogm i just muxed in wmp 6.4, i got an error that said something like "Wrong file version!" twice, and then the movie played as normal. playing it back in bsplayer caused a crash. an ogm i made a few days ago seemed to work fine though. deleting the new dlls and reinstalling the oggds filter and its dlls fixed the problem. just so you know that there may be some issues with oggds not seeing the dlls it expects.

i agree with you matter, i am sick of dlls :) i think i'm going to just stick with the new ICL headac3he one and the 0.9.8.6 oggds unless tobias releases a new stable, fast version of his filters. this technology is new enough anyway without me swapping stuff around further creating more bugs :)

[MaTTeR]

The muxing problem is exactly what I was afraid of with replacing what was in the sys32 dir. I like the libvorbis.dll idea better which could go into the BeSweet dir.

Thx for the heads up!

[DSPguru]

if you don't want to change the windows/systems dlls, you should try keep the tracking on libVorbis.dll (http://www.inf.ufpr.br/~rja00/files/libvorbis.zip).
again, this isn't compiled with ICL, but it's updated almost daily.

anyway,
i might create an ICL of libVorbis.dll in the future, but in case Tobias will release his dlls, it would be redundant. so i'll wait and see.

[DarkAvenger]

@canadian_fbi

Hmm, strange maybe I got my sources messed up.. The MSVC6 dll is new, could you check filesize with it?

[canadian_fbi]

the MSVC6 dll showed up as "vorbis.dll 1.11 (18.03.2002) engine 1.66" in the log, and had the same filesize and last modification date as the 3/18 one, so i assumed it was the same as the old one. i just checked the filesizes now and they're both the same - still oversized.

[DarkAvenger]

I think I did some crap, could you try again? Wait..Ok.

[DSPguru]

Tobias has released his new filter (http://tobias.everwicked.com/oggds.htm) :Version 0.9.8.9

o There was a bug in the memory management which caused sometimes a crash of the splitter or multiplexer
o A flag in the media tiype was not set correctly. Therefore all frames were marked as key frames when multiplexing to AVI
o I'm using now the intel compiler (ICL). This reduces specially the vorbis encode time dramatically. (On my PC with Athlon 1.4GHz to 50%!)
Thanks to Stefan for this hint. so you might wanna check the effect on the OggMachine.

Dg.

[canadian_fbi]

@darkavenger: the new msvc6 compile you've put up on your site seems to fix the problem i had. it gave a filesize of 3934 KB at a time of :58 in dumb mode (compared to the results posted earlier).

@dspguru: the new dlls give a good speed boost, just like the ICL ones did for headac3he. oggmachine encoding is down to :56 now. hopefully the 0.9.8.9 oggds filters will be a little more stable so one can take advantage of the speed boost.

much thanks to everyone who's helped clear up this problem in such a fast amount of time. a 25% speed boost in the time since i first reported that oversizing bug is huge (even if it is mostly an improvement by the vorbis guys). you really are putting out some fantastic programs in a short period of tiem.

[DarkAvenger]

Thx canadian_fbi. Indeed I made crap as I messed up the sources. I forgot to update some .h files... Shame on me. :rolleyes: But good to know that now everything is alright. (And good that Roberto did not make the same mistake...)