If I understand him correctly, most of his tunings are parameters modification of noise masking threshold.
As found in GT3, at >q3, this one produces obviously bigger files compared with official 1.0.1.
from README
OggEnc modest tuning version beta 1.
This is my personal (experimental) tuning of OggEnc against
HF boost problem. Please use this binary only for evaluation,
not for archiving. I don't know this tuning performs well or not,
but I attach a patch to 1.0.1 source code in the hope that
the patch will help those who have interested in OggEnc tuning,
even if this tuning ends up in failure.
-q0...-q2: not tuned.
-q3: around 128kbps
-q4: around 160kbps (or higher)
-q5: around 192kbps, recommended
-q6: around 230kbps
-q7: around 260kbps
-q8...-q10:
nyaochi
http://nyaochi.cocolog-nifty.com/audio/200...c_modest_t.html (http://nyaochi.cocolog-nifty.com/audio/2003/12/oggenc_modest_t.html) (in Japanese language)
oggencmtb1 binary (http://homepage3.nifty.com/nyaochi/soft/dist/oggencmtb1.zip) for MS-Windows
If I understand him correctly, most of his tunings are parameters modification of noise masking threshold.
As found in GT3, at >q3, this one produces obviously bigger files compared with official 1.0.1.
from README
OggEnc modest tuning version beta 1.
This is my personal (experimental) tuning of OggEnc against
HF boost problem. Please use this binary only for evaluation,
not for archiving. I don't know this tuning performs well or not,
but I attach a patch to 1.0.1 source code in the hope that
the patch will help those who have interested in OggEnc tuning,
even if this tuning ends up in failure.
-q0...-q2: not tuned.
-q3: around 128kbps
-q4: around 160kbps (or higher)
-q5: around 192kbps, recommended
-q6: around 230kbps
-q7: around 260kbps
-q8...-q10:
nyaochi
http://nyaochi.cocolog-nifty.com/audio/200...c_modest_t.html (http://nyaochi.cocolog-nifty.com/audio/2003/12/oggenc_modest_t.html) (in Japanese language)
oggencmtb1 binary (http://homepage3.nifty.com/nyaochi/soft/dist/oggencmtb1.zip) for MS-Windows
Nice. I wonder when guruboolez will be available to test it out.
I did a quick abx on the Waiting sample at -q 3 (average 142.6 kbps). Rather than listening for other types of distortion, I focused on HF boost (stronger hiss) in the range 14.6 to 16 secs. I guess for pretty average ears, I didnt do so bad.
WinABX v0.23 test report
12/29/2003 12:20:30
A file: D:\listeningTests\Waiting.wav
B file: D:\listeningTests\ogg.wav
12:23:43 0/1 p=100.0%
12:23:58 1/2 p=75.0%
12:24:20 2/3 p=50.0%
12:24:42 3/4 p=31.2%
12:24:56 4/5 p=18.8%
12:25:11 5/6 p=10.9%
12:25:22 5/7 p=22.7%
12:25:43 6/8 p=14.5%
12:26:02 6/9 p=25.4%
12:26:26 7/10 p=17.2%
12:27:08 test finished
Just tested on two samples.
The first one on strings (pizzicato) : 8/8
The second one on cello : 8/8
On the second file, even annoying artifacts (boiling sound) was lowered with this patched version (I'll upload the short part soon). On the other side, bitrate was a bit higher between CVS -q 4 and patched -q 3 (+ 10 kbps).
EDIT : sample available here (http://www.hydrogenaudio.org/forums/index.php?showtopic=14134&st=35&#).
I tried again, concentrating a bit more. The HF boost is still there in q 3 at least:
WinABX v0.23 test report
12/29/2003 12:30:01
A file: D:\listeningTests\Waiting.wav
B file: D:\listeningTests\ogg.wav
12:31:56 1/1 p=50.0%
12:32:33 2/2 p=25.0%
12:32:56 3/3 p=12.5%
12:33:08 4/4 p= 6.2%
12:33:28 5/5 p= 3.1%
12:36:41 6/6 p= 1.6%
12:37:21 6/7 p= 6.2%
12:37:41 7/8 p= 3.5%
12:38:06 8/9 p= 2.0%
12:38:40 9/10 p= 1.1%
12:38:47 test finished
Tried on an organ sample : difference in favour of patched version was easy to detect. Nevertheless, difference between original file and patched vorbis was easier...
It's a step in the good direction (of course if overall quality is the same), but some additional work is needed before having a noise-free encoder.
The really good thing is that he's provided a patch file which shows all the things he changed. I had a brief read of it and now have a fairly good idea which parameters are changed in tuning vorbis.
The really good thing is that he's provided a patch file which shows all the things he changed. I had a brief read of it and now have a fairly good idea which parameters are changed in tuning vorbis.
It sounds nice. Do you think that some of these changes may introduce some artifacts? Because it's one thing to notice the decrease in noise, it's another to be sure that other parts of the signal are affected in a bad way. Anyway, I hope that significant changes will be soonly introduce in official vorbis encoder.
N.B. That's funny, during the same week, we saw two different attempts of single developer for improving vorbis encoder More AFAIK than the 18 last months !
I know he doesn't stay on pre-echo as target. I tried ABX on castanets.wav at -q 5(233kbps) though.
mtb1:15/15 better than 1.0.1(-q 5, 171kbps), slightly worse than GT3b1(-q 5, 215kbps)
i searched and couldn't find sources...
QK: you said he had a patch out (against 1.0.1 i presume) ?
could you point me in the direction?
thanx
i searched and couldn't find sources...
QK: you said he had a patch out (against 1.0.1 i presume) ?
could you point me in the direction?
thanx
http://homepage3.nifty.com/nyaochi/soft/dist/oggencmtb2.zip (http://homepage3.nifty.com/nyaochi/soft/dist/oggencmtb2.zip)
ah. thanx QK
i assumed that this was just an oggenc win32 binary, but it includes source also!
eheh...
later
ah. thanx QK
i assumed that this was just an oggenc win32 binary, but it includes source also!
eheh...
later
xmixahlx,
I made a mistake in making the patch in the zip file. I want to take back the source code, which does not exist in my hard disc. Since I remember MTb2 was more simplified tuning than MTb1, I'm going to rewrite the tuning. So please don't use the source code to make a binary. Thank you.
nyaochi - are you meaning that the patch in the zip file is incomplete? --> because it (modest tuning beta 2) does apply to 1.0.1 source, and it does build + seems to work properly (from minimal testing...)
i've already made libraries based on the patch and compiled oggenc dynamically against them:
http://rarewares.org/debian.html (http://rarewares.org/debian.html)
if there is an error i'll take them offline
later
nyaochi - are you meaning that the patch in the zip file is incomplete? --> because it (modest tuning beta 2) does apply to 1.0.1 source, and it does build + seems to work properly (from minimal testing...)
i've already made libraries based on the patch and compiled oggenc dynamically against them:
http://rarewares.org/debian.html (http://rarewares.org/debian.html)
if there is an error i'll take them offline
later
Yes. I made a mistake including another version of patch (maybe post-beta2 for my private test) in that archive. It should work properly and show similar results as MTb2 judging from the content of the patch, but its behavior will be slightly different from MTb2.
Anyway, I have just released MTb3. I tuned around 128kbps a little and remapped (corrected) quality-bitrate mapping in order to coordinate with the official encoders. We had to set quality 3 to get 128kbps with MTb2 due to the bitrate inflation, but now, we can set quality 4 to get 128kbps with MTb3, which is the normal behavior of the official or other tuned encoders. I believe MTb3 is easy-to-use for everyone without quality regression from MTb2.
OggEnc Win32 binary and patch is available here (http://homepage3.nifty.com/nyaochi/soft/dist/oggencmtb3.zip). So could you please re-build libvorbis with this patch? Thanks in advance.
Anyway, I have just released MTb3. I tuned around 128kbps
excellent!
So could you please re-build libvorbis with this patch? Thanks in advance.
absolutely. new builds in a few minutes
later