There is a new aoTuv beta release(4.5) out.

Changelog:


http://www.geocities.jp/aoyoume/aotuv/index.html

Let's call together: Gu-ru-boo-lez!!!

Erm.. as if Guru had big interest in low bitrate testing

Come on people! This is low bitrates area, you could test it yourself...

Thanks, i tried it...but sounds like metallic, the switch used is:
"-q-2" the nominal bitrate whas 32.

Any vorbis can encode at 20-25? (at 44 and stereo)

Did you compare it to aoTuV b4?

Is the encoding at high quality affected by this? I encoded a file at -q6 with both b4 and b4.5 and decoded again (with the same decoder), and the content of the decoded files differs (as shown by a quick md5 hashing).

MedO

Can you find out 'audio difference' between them in some audio editor? How much differs the size of encoded files?

He's already busy with Nero AAC.

wow, nice works

I've tried -q 4 , they can be considered bit identical

Beta4 = 3,489,542 bytes
Beta4.5 = 3,489,535 bytes

-q 2
Beta4 = 2,768,456 bytes
Beta4.5 = 2,805,795 bytes

The slight difference in -q4 might be bacause of the beta4 I used is P4 optimised, Optimised version might produce a file that has slight differences from non-optimised version.....this causes discrepencies in MD5, but it terms of perception quality, it is the same.

Lets wait a bit, maybe Aoyumi will give some explanations about this version and what we should expect from it. If it is OK, it would be wonderful if john33 compiles an OggDropXPd with this thingy inside. Is it possible, John ?

I will be testing this new release as well. I have high expectations about it, but aside the tuning at very low bitrates I don't suppose it's a major release, probably why it's numbered 4.5 and not 5.0

Aoyumi-san domo arigato

Ah, I wonder if one day you will try making the first known implementation of Bitrate Peeling. I know I would be happy about just better tuning at the higher bitrates, so that Vorbis could finally beat Musepack.

I use -q-2 sometimes on classical stuff for some reason. I have a symphony of 63 Minutes, which encodes @ 13.960.454 Bytes. With 32 kbps it does sound incredible!

Now I encoded it again with ao4.5.

I could not hear any differences/improvements at the problematic parts.

BUT theres a significant improvement in size:

aotuv4: 13.960.454 Bytes
aotuv4.5: 13.080.681 Bytes

Cool, I'll give this a try this evening since I use Vorbis at -q2.25 on my iFP-895 (with a hacked firmware to play Vorbis files that average less than 96 kbs.)

Where can we find the firmware ?

http://www.misticriver.net/showthread.php?t=24020

For the moment, it is impossible.


Please hear and judge in person.

I downloaded this and compiled it. After unpacking the .tar.gz I had to do chmod +x on the configure and install-sh scripts before building and installing it. Other than that, no problems.

I encoded a few files and in general they are free of artifacts, except one: The encoder seems to have a strong preference for removing high frequencies at low quality settings. The encoded files sound as if they were downsampled or a low-pass filter was used. This effect diminishes with higher quality settings, and above -q7 I can't tell the difference anymore. At lower settings I can distinguish the vorbis file from the original because of the relatively increased bass and decreased treble.

I am just curious if this was intentional in the design of the encoder? I do think it is less annoying than other types of artifacts, so maybe it's not such a bad trade-off.

I can´t really hear the diference, is hard to say that is a big improvement, ok I only made one test (at -q 0), so I can´t really say that is not effective and I don´t have a "extreme hear" so it just only my opinion.

Yes, it's possible now I have the source, which wasn't available when I first looked.

Extract from one of the text files included:

Hi, n00b here. Downloaded, chmod'ed, configured & compiled... no errors.
But where is the executable ? Or do I miss something here ?

EDIT: Yes I did
It was libvorbis source. Oggenc source downloadpage gives me Error 404.

Here is what I did:

Download libogg-1.1.2.tar.gz and vorbis-tools-1.1.1.tar.gz from http://www.xiph.org/downloads/
Download libvorbis-aotuv_b4.5.tar.gz.tgz from http://www.geocities.jp/aoyoume/aotuv/
Unpack everything.
cd libogg-1.1.2
./configure
make
make install
cd ../aotuv-b4.5_20051105
chmod +x configure
chmod +x install-sh
./configure
make
make install
cd ../vorbis-tools-1.1.1
./configure
make
Run oggenc/oggenc

Which file was bigger? aoTuV b4 or aoTuV 4.5? May be this improvment affects size while keeping quality same?
By the way, did you use one of the famous 'hard samples' or any randomly chosen 'easy sample'?

I know this is not reccomended, but with oggdropXPd (and I think also with other programs) is possible to go down to 12 Kbps @ 44.1 KHz stereo using ABR with:
Min. Bitrate = 0
Nom. Bitrate = 12
Max. Bitrate = 0

I think Min. Bitrate and Max. Bitrate = 0 means there is no limit, thus give a higer quality than CBR @ 12 Kbps.

Sorry, I just found that ABR @ 12 Kbps with oggdropXPd is automatically resampled to 8 KHz, while ABR @ 32 Kbps is resampled to 24 KHz.

Generic, P3 and P4 compiles of oggdropXPd now at Rarewares.

Thanks John...

Could you update also libvorbis.dll with this new Aoyumi stuff ?

Surely, but do people want the full range of compiles, or is this considered somewhat experimental at this stage?

I didn't test it thoroughly, but it seems to me that the most important point was to increase the effectiveness of encoder by making same samples sound the same at lower bitrate. For example, I cannot distinguish one sample @ -q2 from another, but the one created with b4.5 is smaller. That's all I can say for now.
I really think that we must gather and do a couple of short tests just to show that b4.5 is better than 1.1, and then make it the recommended version. Aoyumi's work deserves it.

Ok, the file created by 4.5 is smaller by a few kbytes, probably this is the improvment, same quality in smaller size... and not is not one of the famous hard samples, is a sample that I choose and probably an easy one...

Anyway great job Aoyumi, all Vorbis fans apreciate your hard work, I imagine that if wasn´t for you, we are still using the 1.0.1 version (or probablily 1.0.2 ) I think that the Xiph boys are busy in other things... like Theora...

Tried it on two samples - harpsicord and Prodigy "You'll Be Under My Wheels" (both of them require high bitrates with b4 q6. 1st one q6 average is 237 and second q6 average bitrate is 300!)

I tested them on -q -2, q 0, -q 1. b4 vs b4.5. I can't say I found out any difference. May be, JUST MAY BE, b 4.5 preserves high frequencies better but I can't be sure.
Also I tested each version against original file on -q 0 and -q 1. And it was slightly more difficult to ABX between b4.5 and original than b4 and original due mentioned possible better high freq preservation.

P. S. Harpsicord sample is not the sample others use. It is just from my collection.

I tried to ABX b4 to original and b4.5 to original in a pop sample at -q0, and I think it was harder with b4.5 when concentrating on the distortions in the main voice only. However, it was still easy when listening for a different artifact (close-to-noise parts of the sound being replaced by real noise... don't know if that makes any sense to you, I just don't know the right terminology) that is present in both.

MedO

Just a quick question here: after comparing b4 to the original and b4.5 to the original but not b4 to b4.5, is it really correct to draw any conclusions about the quality improvements from b4 to b4.5? I would call the method above a sighted test, and would only use that when the differences are very obvious... Were there obvious differences in your files between b4 and b4.5?

In my test the difference was subtle. I think if we want reliable test, a listener first has to find such -q meaning (with potentially weaker encoder version) at which ABXing becomes VERY DIFFICULT. Then use the new version and encode a file which is about the same size as 1st encoding. If there are any improvements - the second encoding will not be possible to ABX. The main difficulty is an ABXer has to find a proper -q meaning for each sample. The second difficulty are improvements which affect quality at lowest bitrates and ABXing from the original becomes very easy. So maybe we should use 'very easy samples'?

The obvious thing to do is a blind test between two encoded files, A=v4.0 and B=v4.5, which have comparable filesizes. It doesn't matter what kind of samples you use. If you can hear a difference between the two versions, then go ahead and rate which one you find more pleasing...

And John please compile oggenc2.6 and ogg vorbis dlls using aoTuV b4.5 as well.

wow, this is amazing at q0, makes me with i had an vorbis portable, i am stuck with using q25 wma...

As I said before the difference between b4 and b4.5 is very small and I can't decide which one is better at such a low bitrate .
I hope others would.

Сool ! Thanks mate!

BTW Aoyumi, are you going to make tunings for high bitrate in the future versions ? Thanks for your great work.

Cheers

I think the question is which one makes the file smaller.

well...just libvorbis.dll is enough for testing I think. Not the full range of compiles.


I asked this because I'm not able to compile myself this library (only vorbisenc/vorbisfile/vorbis in fact w/o optimizations).

With my samples sizes are almost identical.

Uhm, waiting for a new Lancer !



B-1B Lancer info page

And when you are using oggenc from vorbis-tools 1.1.1, how does this use aotuv4.5 ?

Because oggenc uses the libvorbisenc/libvorbisfile libraries to do the encoding. So as long as you replace this library you should get the new encoding functionality.

You can also test it without installing and breaking packages:


oggenc will now temporarily link against the new libraries and will encode with aotuv:

yeah it's pretty handy for packaging

from RareWares/Debian:

btw, aotuv 4.5 is at RW/Debian now


later

I think aotuv beta 4.5 is safe to use.
my logic: new Aoyumi releaze is BETA 4.5, not new TEST version.

So, b4.5 is supposed to perform better than b4 @ q3 and below, but what about @ q3 and better?

Sounds fine to me . 4 and 5 were transparent to me long before all of the tweaking with the psychoacoustics. The only major thing that really needed to corrected was some of the noise normalization issues, which are still being addressed and have been fixed. How much better can you make it possible? Aoyumi has been at it for a while now and has done a great job.