IPB

Welcome Guest ( Log In | Register )

> Hydrogenaudio Forum Rules

- No Warez. This includes warez links, cracks and/or requests for help in getting illegal software or copyrighted music tracks!
- No Spamming or Trolling on the boards, this includes useless posts, trying to only increase post count or trying to deliberately create a flame war.
- No Hateful or Disrespectful posts. This includes: bashing, name-calling or insults directed at a board member.
- Click here for complete Hydrogenaudio Terms of Service

4 Pages V   1 2 3 > »   
Reply to this topicStart new topic
Is LAME capable of calamitous MP3 quality ?, ...or is there a (worse) alternative
NumLOCK
post Nov 21 2002, 20:30
Post #1


Neutrino G-RSA developer


Group: Developer
Posts: 852
Joined: 8-May 02
From: Geneva
Member No.: 2002



Hi all,

I'd like to generate a very bad sounding MP3 file, in order to expose as many obvious artifacts as possible. ohmy.gif

Also, it would be nice to have a bitrate that's as high as possible (256 or 320kbps). blink.gif

I've already played with the following options a tiny bit (using LAME 3.92):

LAME.EXE Test.wav
-ms => full stereo, for a nice bitrate waste wink.gif
-q 9 => worst possible noise shaping
--athonly => disable psychoacoustic model altogether - except ATH
--noshort => optimize for maximal transient smearing ph34r.gif
--nores => disable bit reservoir to decrease quality on difficult to encode parts B)
-k => disable all filters, for nice ringing and aliasing artifacts ohmy.gif
--strictly-enforce-ISO => should cripple the output file quite a bit.

The sound now has several clear artifacts @ 128kbps, but that's not obvious enough... How would you suggest to decrease the quality even more, while increasing the bitrate ? It should be bad-sounding, even for someone who hasn't heard the original.

I thought about a few "improvements":
=> use short blocks all the way, in order to kill coding efficiency, and to lose most advantages of transform coding laugh.gif
=> re-enable the psymodel - and break it using cmdline-options only.

Transcoding from a 20kbps WMA is not an option... I'd like it to be a pure printine MP3 encode, from the original CD and with an unmodified mp3 encoder.

By the way, it's just for a fun experiment, to see how far we can go.
I love LAME, and no harm is intended !!! BTW, if you think the crime could be commited with a better tool [blade/xing_old/etc] please let me know.

What would you suggest?
Thanks

This post has been edited by NumLOCK: Nov 21 2002, 22:30


--------------------
Try Leeloo Chat at http://leeloo.webhop.net
Go to the top of the page
+Quote Post
Ivan Dimkovic
post Nov 21 2002, 20:35
Post #2


Nero MPEG4 developer


Group: Developer
Posts: 1466
Joined: 22-September 01
Member No.: 8



Try velvet.wav (from mp3dev.org) with Blade @128 smile.gif It will fit your purpose smile.gif


Or ask MS - what MP3 encoding setup they use when comparing MP3 to their coding solution smile.gif

This post has been edited by Ivan Dimkovic: Nov 21 2002, 20:37
Go to the top of the page
+Quote Post
NumLOCK
post Nov 21 2002, 20:52
Post #3


Neutrino G-RSA developer


Group: Developer
Posts: 852
Joined: 8-May 02
From: Geneva
Member No.: 2002



Ivan,

Wow ! what a fast response..

Thank you for the tip, I've just tried it - it sounds nasty but not quite enough..

After a quick comparison, on this sample Blade is (for me) a tiny bit higher quality than with "tweaked" LAME as above :-(

Actually I'd need to make the encoder sound awful on pretty much anything.. and since there doesn't seem to be room for tweaking Blade..

As for MS, that's a clever advice but I usually don't deal with them ph34r.gif

This post has been edited by NumLOCK: Nov 21 2002, 20:55


--------------------
Try Leeloo Chat at http://leeloo.webhop.net
Go to the top of the page
+Quote Post
guruboolez
post Nov 21 2002, 21:00
Post #4





Group: Members (Donating)
Posts: 3474
Joined: 7-November 01
From: Strasbourg (France)
Member No.: 420



Plugger is an unbeatable codec. Even JuraXing is better.
Try Plugger :

http://ec2000.xperiment.net/
Direct link
Go to the top of the page
+Quote Post
NumLOCK
post Nov 21 2002, 21:17
Post #5


Neutrino G-RSA developer


Group: Developer
Posts: 852
Joined: 8-May 02
From: Geneva
Member No.: 2002



Thanks for the link Guruboolez !
I must admit, that Plugger codec gives quite amazing treble @ 128kbps with Velvet.wav biggrin.gif

However after trying it @320kbps I'd say it doesn't sound shockingly bad (for the average person of course).. well, except with the fatboy sample. blink.gif

So far, the worst 128k setup (for me) is:
LAME 3.92 -ms -q 9 --athonly --noath --noshort --notemp --nores -k --strictly-enforce-ISO

This is impressive, but not quite sufficient for an increase of bitrate yet...
Maybe someday, I'll try to get back to the origins (2.x?) of LAME, gentlemen laugh.gif

This post has been edited by NumLOCK: Nov 21 2002, 21:56


--------------------
Try Leeloo Chat at http://leeloo.webhop.net
Go to the top of the page
+Quote Post
Jens Rex
post Nov 21 2002, 21:57
Post #6





Group: Super Moderator
Posts: 605
Joined: 18-December 01
Member No.: 680



How about -p for a nice waste of bits for CRC?

I love this thread biggrin.gif.
Go to the top of the page
+Quote Post
Ivan Dimkovic
post Nov 21 2002, 22:09
Post #7


Nero MPEG4 developer


Group: Developer
Posts: 1466
Joined: 22-September 01
Member No.: 8



QUOTE
LAME 3.92 -ms -q 9 --athonly --noath --noshort --notemp --nores -k --strictly-enforce-ISO


Hmm.. try resampling to 48000 as well smile.gif Why not waste extra bits/sample smile.gif

Also, is there an option to use dual-channel independent bit allocation (dual mono)? that way the allocation will be fixed to 64 kbps per channel, without a chance to donate bits or to exploit intra-channel redundancy.

Also,

-m i intensity stereo - maybe without psychoacoustic IS can make even more horrible artifacts than dual mono, worth trying smile.gif
Go to the top of the page
+Quote Post
NumLOCK
post Nov 21 2002, 22:46
Post #8


Neutrino G-RSA developer


Group: Developer
Posts: 852
Joined: 8-May 02
From: Geneva
Member No.: 2002



Thank you Ivan !!
Sampling rate bloating.. What a nice way to decrease storage efficiency by almost 10% :-)

I have included the useless CRC calculation from JensRex also.

The new command line is:
lame.exe --resample 48 -ms -q 9 -p --athonly --noath --noshort --notemp --nores -k --strictly-enforce-ISO

Isn't there a way to disable the %@&"~/&* huffman coding ? I mean, with this thing were disabled or at least seriously broken, we could pack more artifacts per kilobyte..

By the way Ivan, I can't seem to enable Intensity stereo with -mi... that's too bad - this option could really do wonders.


--------------------
Try Leeloo Chat at http://leeloo.webhop.net
Go to the top of the page
+Quote Post
ssamadhi97
post Nov 21 2002, 22:58
Post #9





Group: Developer (Donating)
Posts: 1203
Joined: 10-February 02
From: Endless Water
Member No.: 1305



QUOTE (NumLOCK @ Nov 21 2002 - 10:46 PM)
Isn't there a way to disable the %@&"~/&* huffman coding ? I mean, with this thing were disabled or at least seriously broken, we could pack more artifacts per kilobyte..

not without entirely breaking compatibility with mp3 decoders tongue.gif


kaps = kilo-artifacts per second? ph34r.gif


--------------------
A riddle is a short sword attached to the next 2000 years.
Go to the top of the page
+Quote Post
Q!
post Nov 21 2002, 23:05
Post #10





Group: Members
Posts: 98
Joined: 26-August 02
From: PL
Member No.: 3199



QUOTE (NumLOCK @ Nov 21 2002 - 11:46 PM)
The new command line is:
lame.exe --resample 48 -ms -q 9 -p --athonly --noath --noshort --notemp --nores -k --strictly-enforce-ISO

I think this line should be added to list of recommended lame settings. smile.gif


--------------------
http://qisgod.host.sk
Go to the top of the page
+Quote Post
NumLOCK
post Nov 21 2002, 23:07
Post #11


Neutrino G-RSA developer


Group: Developer
Posts: 852
Joined: 8-May 02
From: Geneva
Member No.: 2002



QUOTE
not without entirely breaking compatibility with mp3 decoders tongue.gif

Oh... you mean "those". You're right, we must keep the huffman codin... but wait !! They could also be used to INCREASE the file size couldn't they?

Too bad they'd bring that benefit losslessly though.. that would have allowed us to go 320kbps while still decreasing quality ph34r.gif

Well, you don't have nothing for less than nothing I guess.

KAPS ? Excellent ! The KAPS and NSR (see below) are now an official measure of sound crapness biggrin.gif

QUOTE
I think this line should be added to list of recommended lame settings.

Agreed. After all, I've just developed the command-line with the highest NSR(*) !!

Also, the advantage with these settings, is that transients are no harder to handle than noise... since both produce artifacted noise at the output.

(*) Noise/Signal ratio. ph34r.gif

This post has been edited by NumLOCK: Nov 21 2002, 23:16


--------------------
Try Leeloo Chat at http://leeloo.webhop.net
Go to the top of the page
+Quote Post
h
post Nov 21 2002, 23:14
Post #12





Group: Members
Posts: 61
Joined: 30-September 01
Member No.: 79



If MP3 can handle custom huffman codebooks, you could make catastrophically poor ones that decrease coding performance by around 800% I would imagine (if the zero-run codes can be changed). A custom tool would be needed of course, or if LAME already creates codebooks internally, the frequency sort could be inverted so the longest codes go the the most common symbols.

-h
Go to the top of the page
+Quote Post
NumLOCK
post Nov 21 2002, 23:24
Post #13


Neutrino G-RSA developer


Group: Developer
Posts: 852
Joined: 8-May 02
From: Geneva
Member No.: 2002



QUOTE
you could make catastrophically poor ones that decrease coding performance by around 800%

oOOOoOOh... 800% better NSR for the same bitrate... this sounds so beautiful to me !
How come no genius had implemented this wonderful idea yet ?!??

Just one thing, we should be careful when choosing the mp3 side-information, because 32-bit addressing might be an issue while huffman coding them... biggrin.gif

Especially the CRC.. this will be the most difficult part to losslessly encode !

Also, remember that the requirement is, at least to be able to code silence at 320kbps... of course, we could switch to freeformat if a difficult-to-encode passage (ie. one or more non-zero samples) is to be encoded.

If the inherent limitations of the MP3 format (ie. bitrate ceiling too low for encoding silence mad.gif ) indeed proves to be an issue, well... I guess I'll develop my own format with highly efficient 2^256-bit huffman codes.. ph34r.gif any takers? biggrin.gif

This post has been edited by NumLOCK: Nov 21 2002, 23:35


--------------------
Try Leeloo Chat at http://leeloo.webhop.net
Go to the top of the page
+Quote Post
niktheblak
post Nov 22 2002, 00:01
Post #14





Group: Members (Donating)
Posts: 302
Joined: 3-October 01
From: Finland
Member No.: 188



I think a really good way of increasing the bitrate while maintaining poorest possible quality (constant ka/s) would be, in addition to the already presented excellent command lines, by activating a dubious VBR mode. Something like -V 7 or -V 8 should do the trick tongue.gif

The problem is that it possibly might sound better than CBR, although I seriously doubt it. Just remember to use a decent bit-eating sample, and you're all set smile.gif

The modified Huffman tables idea is pretty good, but I consider it cheating. It has to sound like crap, fair and square smile.gif

Edit:

You also could normalize the sample to 110% or so in order to get some serious clipping...

Edit2:

Nah, artificially introducing clipping would also be cheating. I also thought of using the experimental switches, why not all of them at once? -X -Y -Z should really wreck some havoc.

Also a really simple way to discover ultra-tuned settings is to look around discussion boards for anything that has the magic words "mY LaMeLiNe r0xX, s0uNDs beTTer tHan aps!!!!!1111". The Newbie settings are usually extremely tuned for artifacts.

This post has been edited by niktheblak: Nov 22 2002, 00:21
Go to the top of the page
+Quote Post
ViPER1313
post Nov 22 2002, 01:26
Post #15





Group: Members
Posts: 271
Joined: 19-August 02
From: Maryland
Member No.: 3109



Instead of using -m s for middle stereo, you can try to use -m d for dual stereo. 2 channels with no coupling / channel distribution might help out a bit. Also, try setting an extremely high low pass filter, such as --lowpass 24. This is too much for 128 to handle. Have fun tongue.gif

PS - If you want to cheat, just set either --lowpass 3 OR --highpass .5 (thats right - .5 - its ALL you need B) )
--highpass .5 will give you the EXACT effect you want

This post has been edited by ViPER1313: Nov 22 2002, 01:39
Go to the top of the page
+Quote Post
kennedyb4
post Nov 22 2002, 03:04
Post #16





Group: Members
Posts: 772
Joined: 3-October 01
Member No.: 180



Does --scale allow you to drive the files into clipping? That would just have to sound like shit. biggrin.gif

Too bad there is no --allshort switch. Or if there was an --ath raise function, you could jack the noise floor up until it sounded as bad as you want.
Go to the top of the page
+Quote Post
Shiki
post Nov 22 2002, 10:36
Post #17





Group: Members
Posts: 79
Joined: 13-March 02
Member No.: 1509



QUOTE (guruboolez @ Nov 21 2002 - 08:00 PM)
Plugger is an unbeatable codec. Even JuraXing is better.
Try Plugger :

http://ec2000.xperiment.net/
Direct link

Damn right! I have a couple of mp3s encoded with Plugger at 128kbps Stereo and they sound simply wonderful! biggrin.gif
Go to the top of the page
+Quote Post
Ivan Dimkovic
post Nov 22 2002, 11:13
Post #18


Nero MPEG4 developer


Group: Developer
Posts: 1466
Joined: 22-September 01
Member No.: 8



Try decreasing ATH with negative number of dB - that should make sound even more beutiful smile.gif

Also, dual mono stereo coding mode should cripple file a little bit further
Go to the top of the page
+Quote Post
Gabriel
post Nov 22 2002, 11:35
Post #19


LAME developer


Group: Developer
Posts: 2950
Joined: 1-October 01
From: Nanterre, France
Member No.: 138



I think that you should try Shine...
Go to the top of the page
+Quote Post
sven_Bent
post Nov 22 2002, 13:26
Post #20





Group: Members
Posts: 691
Joined: 15-December 01
From: Denmark
Member No.: 655



biggrin.gif

Damn my stomach hurts now

laugh.gif


--------------------
Sven Bent - Denmark
Go to the top of the page
+Quote Post
NumLOCK
post Nov 22 2002, 13:26
Post #21


Neutrino G-RSA developer


Group: Developer
Posts: 852
Joined: 8-May 02
From: Geneva
Member No.: 2002



People.. I didn't expect such strong support ! From the bottom of my heart I'm thanking you (don't know if my ears agree though).

QUOTE
I think a really good way of increasing the bitrate while maintaining poorest possible quality (constant ka/s) would be, in addition to the already presented excellent command lines, by activating a dubious VBR mode. Something like -V 7 or -V 8 should do the trick

The problem is that it possibly might sound better than CBR, although I seriously doubt it. Just remember to use a decent bit-eating sample, and you're all set

Well, thanks for the suggestion. I'll try to badly enable VBR this way, and hope there'll be some bitrate increase @ constant ka/s.
For the actual sample - even though fatboy.wav is good to measure ka/s - it will be a Dream Theater song. It still contains some nice attacks.

Edit: Wow ! Even -V9 uses 320kbps almost all the time, because of the crippled psychoacoustics I guess. So, we should find a setting where psychoacoustics is enabled, but fails.

QUOTE
The modified Huffman tables idea is pretty good, but I consider it cheating. It has to sound like crap, fair and square

Yes, I agree. To get those artifacts, we've got to earn'em fairly.

QUOTE
You also could normalize the sample to 110% or so in order to get some serious clipping...
Edit2:
Nah, artificially introducing clipping would also be cheating. I also thought of using the experimental switches, why not all of them at once? -X -Y -Z should really wreck some havoc.

Well, that's what WMA actually does when encoding.. so I think maybe, we might want to consider this, to get as many advantages of WMA as possible. Does someone know what the WMA scaling factor actually is ?

QUOTE
Instead of using -m s for middle stereo, you can try to use -m d for dual stereo. 2 channels with no coupling / channel distribution might help out a bit. Also, try setting an extremely high low pass filter, such as --lowpass 24. This is too much for 128 to handle. Have fun

PS - If you want to cheat, just set either --lowpass 3 OR --highpass .5 (thats right - .5 - its ALL you need )
--highpass .5 will give you the EXACT effect you want

Oh ! I didn't know there was something better than -ms ! What a sweet hidden setting you found here :-P
This shall be added to the cmdline immediately.
About the lowpass, I've already done the worst possible thing: disabling every filter. You should listen to the real sweet twinkling !
Also, I've tried --lowpass .5 already, and am still considering it.. but maybe that's cheating already.. Hell, isn't full bandwidth, and max ka/s what we want?

QUOTE
Does --scale allow you to drive the files into clipping? That would just have to sound like shit.

Yes.. currently trying this.

QUOTE
Too bad there is no --allshort switch. Or if there was an --ath raise function, you could jack the noise floor up until it sounded as bad as you want.

Actually --allshort does exist. Before enabling dual mono it didn't sound worse than --noshort to me, but now @ 128kbps it might just decrease the quality if we're lucky :-)
Yeah, I've even tried negative values to increase the absolute threshold of hearing but it didn't seem to change anything.. Maybe a LAME developer could give us directions here?

QUOTE
I think that you should try Shine...

Shine? haven't heard of it.. (which might be a good sign). Jumping on Google right away !

Updated cmd line is:
lame.exe --resample 48 -md -q 9 -p --athonly --noath --noshort --notemp --nores -k --strictly-enforce-ISO

This post has been edited by NumLOCK: Nov 22 2002, 13:38


--------------------
Try Leeloo Chat at http://leeloo.webhop.net
Go to the top of the page
+Quote Post
Gecko
post Nov 22 2002, 14:00
Post #22





Group: Members
Posts: 943
Joined: 15-December 01
From: Germany
Member No.: 662



Please be aware that this lame commandline still sounds better than Plugger on the few samples I tested. Further tweaking is required. Perhaps this level of quality can only be reached via code level modifications.
Go to the top of the page
+Quote Post
kennedyb4
post Nov 22 2002, 14:06
Post #23





Group: Members
Posts: 772
Joined: 3-October 01
Member No.: 180



QUOTE (Gecko @ Nov 22 2002 - 08:00 AM)
Please be aware that this lame commandline still sounds better than Plugger on the few samples I tested. Further tweaking is required. Perhaps this level of quality can only be reached via code level modifications.

That's pretty funny. biggrin.gif
Go to the top of the page
+Quote Post
Gabriel
post Nov 22 2002, 14:14
Post #24


LAME developer


Group: Developer
Posts: 2950
Joined: 1-October 01
From: Nanterre, France
Member No.: 138



You should add -V9 --cbr
Go to the top of the page
+Quote Post
NumLOCK
post Nov 22 2002, 14:22
Post #25


Neutrino G-RSA developer


Group: Developer
Posts: 852
Joined: 8-May 02
From: Geneva
Member No.: 2002



QUOTE
Please be aware that this lame commandline still sounds better than Plugger on the few samples I tested. Further tweaking is required. Perhaps this level of quality can only be reached via code level modifications.

LOL ! Thank you for the information... but code level modifications are out of the question.. it *should* really be possible to get to the needed level of performance just by commandline tweaking.

Here's a new commandline that uses VBR technology; the ka/s unfortunately lowers of fatboy.wav (because of the 320kbps frames) but it is much more interesting on real-life material:

lame.exe -V9 -q9 --resample 48 --interch 1 -md -p --allshort --notemp --nores -k --strictly-enforce-ISO

Is it possible to tweak the --interch parameter further?

This post has been edited by NumLOCK: Nov 22 2002, 14:23


--------------------
Try Leeloo Chat at http://leeloo.webhop.net
Go to the top of the page
+Quote Post

4 Pages V   1 2 3 > » 
Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 



RSS Lo-Fi Version Time is now: 17th September 2014 - 16:40