BeSweet & LAME v3.92

[DSPguru]

BeSweet v3.b11 (besweet.notrace.dk) now comes with two versions of lame_enc.dll v3.92 :
- the official v3.92 (slightly modified)
- a lame_enc.dll based on Alexander Leidinger's nonlinear psymodel


please post your impression from the nonlinear lame_enc.dll !


EDIT : note that v1.3b11 is an upgrade to old versions of BeSweet, so you must have v1.2/v1.3b9 before you can use v1.3b11.

[Rumata]

What's nonlinear lame_enc.dll ? What do I need it for ?

[DSPguru]

BeSweet is using lame_enc.dll in order to create mp3.
simplyfied : lame_enc.dll is compressing audio samples into mp3 by "throwing away the less important information".
how does it decide which information is "important" and which is not ?
it uses a psycho-acoustic model.

the lame_enc-nonlinear.zip inside BeSweet v1.3b11 holds a dll that is based on a new psycho-acoustic model, that should acheive better quality at the same bitrates with 'regular' lame_enc.dll.


try it!
Dg.

[Rumata]

Thank's for the explanation.

So the only way to test it is subjective "hearing test" ?

[DSPguru]

Originally posted by Rumata
So the only way to test it is subjective "hearing test" ? there are other ways to test it, but there's no substitute for human ears. (encoder's target).

[MaTTeR]

I finally got around to doing a quick listening test last night with the nonlinear psymodel version. I immediately noticed the bass tones now sound much deeper and the sub hits a little harder now:)

Is it just me or has anyone else noticed Vorbis clips the bass a little much?

I'm setting up a more complete listening tests later tonight which will also include AAC.

[DSPguru]

Originally posted by MaTTeR
I'm setting up a more complete listening tests later tonight which will also include AAC. now this is a real fun !
maybe we should start a new thread with inputs from users comparing :
0. lame_enc 3.91
1. lame_enc 3.92
2. lame_enc 3.92, nonlinear version
3. aacenc 2.15
4. vorbis rc3
5. (lib)vorbis pre-rc4

[MaTTeR]

Originally posted by DSPguru
[B]now this is a real fun !
maybe we should start a new thread with inputs from users comparing

Well I personally like the idea:) Especially for the up and coming AAC format.
__________________
Edit- Just finished a nice listening test with the new non-linear LAME. My conclusion is different than before. If Vorbis didn't exist and I were forced to use LAME then I'd stay with the standard LAME instead of the non-linear. Comparing the "Replacement Killers" track, my ears always liked the original LAME version better. The bullets seemed to sound more real as they whizzed by my head:)

However the clear overall winner to my ears was AAC. The sound was definitely much richer and not as flat. The bass was deeper and the trebles didn't seem to be clipped as much.

I didn't include Vorbis samples in any of the test. Why? I'm starting to notice the Vorbis bass is a little muffled on my sub and not as true as what MP3 and AAC are providing. Has anyone else noticed this yet?

Let me know if you'd like me to post the rather long log file here.

[Antimon]

I dont nessicarily want to see a log file, but bit rates and final file sizes would be nice :-)

[MaTTeR]

Good point Antimon:D I must be getting very tired.

AAC-
repl.ff.q3.aac: 2,580kb (94kbps)
repl.ff.q2.aac: 3,053kb (80kbps)
repl.ff.q1.aac: 3,699kb (70kbps)
repl.ff.normal.aac 7,923kbps (148kbps)

The first 3 files were trancoded using the quality modes 1,2 and 3. The file with "normal" is using the VBR preset. I definitely liked the normal VBR file over the quality modes but I should also mention that would be more than adequate for a 1 CD rip. Yes I double checked the numbers above and it isn't a typo. Can someone explain the bitrate to filesize issue I'm seeing?

MP3-
cell.bswt.altpre128.mp3: 6,618kb (126kbps)
cell.bswt.altpre128.nolin.mp3: 6,618kb (124kbps)
cell.bswt.altstnd.mp3: 9,871kb (185kbps)
cell.bswt.altstnd.nonlin.mp3: 11,000kb (206kbps)

Note: nonlin in the filename means they were encoded with the newer non-linear psymodel.

[Antimon]

How do i swich between the vorbis 3 rc3 encoder and the pre rc4 daily compiles, i have the daily zip but it has 4 dll's none fothem, libvorbis for besweet.

Well usign a wav from a cd (Paul Oakenfold - Rule The Planet Planet Of The Apes remix) in pro logic mode on my decoder this is what i get... *shooting for around 96 kb/s vorbis was abr 96 and aac .3 i belive, the mp3's were cbr*

Lame 3.92 and 3.92 non linear, damn if i can tell them apart....i think the original one is slightly truer and cleaner but i keep going back and forth, both fall short to oog but by no large margine (2,852 kb for both)

Vorbis (2,714 kb) wins thsi one hands down it has the cleanest stereo/suround clarity and seperation, nearly indistigishable from the wav

The AAC was disapointing, even the left and right chanals were a little indescrete and shimering in and out of proper volume, not by a lot but very noticable.....the surround speakers were worse, constant artifact noise and compltly off base with what should be there comapred to the oog/wav (2,828 kb)

in headphones however, the differnces virtually disapear, even the aac passes for a decent "mp3" it does lack some definition and has soem audable artifacts.

This was all flat EQ using winamp.


l

[MaTTeR]

@Antimon

You used the quality modes when transcoding to AAC? Also you stated flat line EQ was used. Does this mean you had it completely turned off or just on and linear? If it was on, then that's very well could be your artifact issue. Ever tried listening to Vorbis with an EQ? It's not pretty.

I recall reading numerous threads at Hydrogen awhile back about which VBR settings to use. AFAIK most recommended using the VBR Normal profile for optimal results and none really recommended quality based modes.

[Antimon]

The eq was disabled/bipassed...

Thats what i did first normal vbr for the aac but it was making a 192 nominal bitrate file....hardly fair to compare that tp a 96 kb/s oog, and if the aac is going to be 192~ kb/s...i'd be gaining nothing by using it over native 2/0 ac3, which ic an either rip directly or generate myself if needbe

[canadian_fbi]

i have to agree that the small tests i did a while ago with aac gave pretty impressive audio quality at pretty unimpressive bitrate results. for the normal preset, anyway. i forget what preset i used to get 106 kbps for aac, but it wasn't really superior to vorbis at that bitrate. another thing i don't like about aac is the constraints on the presets... i wouldn't mind 1-10 integer quality settings compared to the basically continuous vorbis quality settings, but those supposedly aren't really recommended. so basically we really only have 1-3 settings that would be relevant to our needs.

i don't mean to sound so harsh since i didn't really test as comprehensively as i should have... but i don't think aac is really worth the sort of attention it's been getting recently. and the patent issues surrounding it make it difficult to say that it's destined for rapid or significant improvement. but again, i'm no aac expert.

[MaTTeR]

Originally posted by canadian_fbi
i don't mean to sound so harsh since i didn't really test as comprehensively as i should have... but i don't think aac is really worth the sort of attention it's been getting recently.

Indeed you may be correct on this, the more listening test I perform the more unimpressed I am with the quality at lower bitrates. I at least wanted to give it a fair shot as a valid solution but it's making things really difficult for me:D

[DSPguru]

@antimon
to encode vorbis with rc3, install OggDS.
to encode vorbis with pre-rc4, download libvorbis.dll and put it in the BeSweet directory.
to encode with lame 3.92 nonlinear, replace lame_enc.dll, with the nonlinear version that is supplied within the v1.3b11 package.
to encode with lame 3.92 linear psymodel, use the original dll that comes with BeSweet.

for more technical info of dlls manipulation, page canadian_fbi - he should be an expert by now ;)

[Antimon]

So besweet will use/revert back to the rc3 ds filters when lib vorbis is not present, and if it is in the besweet dir it will use libvorbis over the rc3 ds filters?

Very cool, and handy.

Now if onyl we had a checkbox to switch within the gui hehe but sooner or later we will adobt one as standard i'm sure so not really worth coding now for testing, it's easy enough to flip back and forth.

So the files in the daily compile, vorbis.dll oog.dll vorbisenc.dll what can i do with those, i'm sure they serve some purpose, is that for like oog drop and such?

[Antimon]

I figured it out, it's the dsf filters for windows from tolbas :-)