IPB

Welcome Guest ( Log In | Register )

 
Reply to this topicStart new topic
FhG vs Lame, CBR 128kbps
westgroveg
post Jul 17 2004, 12:01
Post #1





Group: Members
Posts: 1236
Joined: 5-October 01
Member No.: 220



I performed a few ABX tests on random samples from rarewares to test out what was better at CBR 128kbps FhG or LAME. I used CEP 2.0's mp3 encoder rather than MP3Enc 3.1 or fastenc 1.02 -hq, MP3Enc will crash if bandwidth is higher than 16000hz not sure why, fastenc just won't encode higher than that, CEP also doesn't have the Blackbird clicking artifact & I just think that CEP's is better tunned but I have only tested on 2 samples & CEP was better.

QUOTE
ABC/HR Version 1.0, 6 May 2004
Testname: FhG vs Lame

1R = C:\Test_Samples\FhG vs LAME\FhG BlackBird.wav
2R = C:\Test_Samples\FhG vs LAME\LAME BlackBird.wav

---------------------------------------
General Comments:
FhG Encoder: Cool Edit Pro 2.0 (Default Settings)
Max Bandwidth: 15804Hz
CBR Bitrate: 128kbps
Sample Rate: 44100Hz
Codec: High Quality Codec (Slowest)
Mid-Side=Yes
Intensity=Yes

LAME Encoder: 3.96, (Default Settings)
Polyphase lowpass filter, transition band: 17249 Hz - 17782 Hz
Encoding as 44.1 kHz 128 kbps j-stereo MPEG-1 Layer III (11x) qval=3, 128.0 kbps   LR, MS

Cool Edit's FhG slow codec does not suffer from clicks as MP3Enc 3.1 or Fastenc 1.02 -hq
do in this sample however the max bandwidth is too low making it dull & easy to spot.
---------------------------------------
ABX Results:
Original vs C:\Test_Samples\FhG vs LAME\FhG BlackBird.wav
    17 out of 17, pval < 0.001
Original vs C:\Test_Samples\FhG vs LAME\LAME BlackBird.wav
    7 out of 17, pval = 0.834


QUOTE
ABC/HR Version 1.0, 6 May 2004
Testname: FhG vs Lame

1R = C:\Test_Samples\FhG vs LAME\FhG castanets.wav
2R = C:\Test_Samples\FhG vs LAME\LAME castanets.wav

---------------------------------------
General Comments:
Both are very close I suspect the first sample sounded better
---------------------------------------
ABX Results:
Original vs C:\Test_Samples\FhG vs LAME\FhG castanets.wav
    11 out of 16, pval = 0.105
Original vs C:\Test_Samples\FhG vs LAME\LAME castanets.wav
    10 out of 16, pval = 0.227


QUOTE
ABC/HR Version 1.0, 6 May 2004
Testname: FhG vs Lame

1L = C:\Test_Samples\FhG vs LAME\FhG DaFunk.wav
2R = C:\Test_Samples\FhG vs LAME\LAME DaFunk.wav

---------------------------------------
General Comments:
Sample 1, max bandwidth is too low making it dull
---------------------------------------
ABX Results:
Original vs C:\Test_Samples\FhG vs LAME\FhG DaFunk.wav
    23 out of 25, pval < 0.001
Original vs C:\Test_Samples\FhG vs LAME\LAME DaFunk.wav
    13 out of 25, pval = 0.500


QUOTE
ABC/HR Version 1.0, 6 May 2004
Testname: FhG vs Lame

1L = C:\Test_Samples\FhG vs LAME\LAME experiencia.wav
2L = C:\Test_Samples\FhG vs LAME\FhG experiencia.wav

---------------------------------------
General Comments:
Sample 2, max bandwidth is too low making it dull
---------------------------------------
ABX Results:
Original vs C:\Test_Samples\FhG vs LAME\LAME experiencia.wav
    8 out of 16, pval = 0.598
Original vs C:\Test_Samples\FhG vs LAME\FhG experiencia.wav
    16 out of 16, pval < 0.001


QUOTE
ABC/HR Version 1.0, 6 May 2004
Testname: FhG vs Lame

1R = C:\Test_Samples\FhG vs LAME\FhG Velvet.wav
2L = C:\Test_Samples\FhG vs LAME\LAME Velvet.wav

---------------------------------------
General Comments:
Sample 1, max bandwidth is too low making it dull, slight ringing

Sample 2, slightly dull, very slight ringing
---------------------------------------
ABX Results:
Original vs C:\Test_Samples\FhG vs LAME\FhG Velvet.wav
    16 out of 16, pval < 0.001
Original vs C:\Test_Samples\FhG vs LAME\LAME Velvet.wav
    14 out of 16, pval = 0.002


QUOTE
ABC/HR Version 1.0, 6 May 2004
Testname: FhG vs Lame

1L = C:\Test_Samples\FhG vs LAME\LAME Waiting.wav
2L = C:\Test_Samples\FhG vs LAME\FhG Waiting.wav

---------------------------------------
General Comments:
Sample 1 sounds slightly worse than 2
---------------------------------------
ABX Results:
Original vs C:\Test_Samples\FhG vs LAME\LAME Waiting.wav
    16 out of 16, pval < 0.001
Original vs C:\Test_Samples\FhG vs LAME\FhG Waiting.wav
    16 out of 16, pval < 0.001


LAME pretty much kicks ass, LAME keeps a higher bandwidth & still prevents artifacts.

What max bandwidth would I have to set in CEP to match lame's default lowpass?

This post has been edited by westgroveg: Jul 17 2004, 12:30
Go to the top of the page
+Quote Post
Gabriel
post Jul 17 2004, 12:38
Post #2


LAME developer


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



Note: in ABR/CBR, lame is using sfb21 in the same way as VBR with -Y. It means that the lowpass is not constantly the one reported by Lame. Lame will encode freqs in sfb21 only if possible considering the bit budget.
Go to the top of the page
+Quote Post
SirGrey
post Jul 17 2004, 12:48
Post #3





Group: Members
Posts: 241
Joined: 8-February 04
Member No.: 11863



QUOTE
MP3Enc will crash if bandwidth is higher than 16000hz not sure why

Hmmm ?
Strange...
But are you sure to use win95 or win98 ?
mp3enc31 does not work on NT based systems usually (if bitrate is > 128 or bandwidth is not default), this was a confirmed bug by Fhg.
BTW, mp3enc seems to be the best Fhg mp3 encoder...
So, try compare it to Lame, be more interesting.
Later fast Fhg encoder family was used in Roberto test and losse to Lame already smile.gif
Go to the top of the page
+Quote Post
rjamorim
post Jul 17 2004, 18:06
Post #4


Rarewares admin


Group: Members
Posts: 7515
Joined: 30-September 01
From: Brazil
Member No.: 81



QUOTE (SirGrey @ Jul 17 2004, 08:48 AM)
Later fast Fhg encoder family was used in Roberto test and losse to Lame already smile.gif
*


Even I probably wouldn't take the results of that test too seriously.

There have been some grave flaws in the planning phase. Some codecs that would probably perform better at CBR (FhG Fast and iTunes) were tested at VBR, and Xing wasn't tested at it's latest version.


--------------------
Get up-to-date binaries of Lame, AAC, Vorbis and much more at RareWares:
http://www.rarewares.org
Go to the top of the page
+Quote Post
SirGrey
post Jul 17 2004, 21:42
Post #5





Group: Members
Posts: 241
Joined: 8-February 04
Member No.: 11863



QUOTE
There have been some grave flaws in the planning phase.

"Nobody is perfect." (Some like it hot) biggrin.gif
Forgot, that FhG encoder was used in VBR mode there, sorry...
Go to the top of the page
+Quote Post
Bogalvator
post Jul 18 2004, 01:00
Post #6





Group: Members
Posts: 19
Joined: 25-February 04
Member No.: 12250



west, what commandline did you use for Lame?
Go to the top of the page
+Quote Post
westgroveg
post Jul 18 2004, 02:42
Post #7





Group: Members
Posts: 1236
Joined: 5-October 01
Member No.: 220



QUOTE
Hmmm ?
Strange...
But are you sure to use win95 or win98 ?
mp3enc31 does not work on NT based systems usually (if bitrate is > 128 or bandwidth is not default), this was a confirmed bug by Fhg.
BTW, mp3enc seems to be the best Fhg mp3 encoder...
So, try compare it to Lame, be more interesting.
Later fast Fhg encoder family was used in Roberto test and losse to Lame already

Yes I am using Windows XP, I wasn't aware of this bug.

If I do some more ABX tests with MP3Enc I'll test it against fastenc & CEP before LAME because I'm pretty confident it's inferior to CEP.

QUOTE
west, what command line did you use for Lame?

Just default as with CEP, later I might try optimized settings.
Go to the top of the page
+Quote Post
ff123
post Jul 18 2004, 05:03
Post #8


ABC/HR developer, ff123.net admin


Group: Developer (Donating)
Posts: 1396
Joined: 24-September 01
Member No.: 12



QUOTE (westgroveg @ Jul 17 2004, 05:42 PM)
If I do some more ABX tests with MP3Enc I'll test it against fastenc & CEP before LAME because I'm pretty confident it's inferior to CEP.


It depends...

On the bitrate, the samples you choose, and if you're sensitive to high frequency artifacting or not.

If you use mp3enc31 at 128 kbit/s, you'll probably want to override the default lowpass, which is pretty low (I think something like 14.5 kHz or so).

ff123
Go to the top of the page
+Quote Post
SirGrey
post Jul 18 2004, 09:29
Post #9





Group: Members
Posts: 241
Joined: 8-February 04
Member No.: 11863



QUOTE
If I do some more ABX tests with MP3Enc I'll test it against fastenc & CEP before LAME because I'm pretty confident it's inferior to CEP.

Of course, test it by yourself.
And, yes, even FhG themselves recommeded to change lowpass from default to -bw 15995 (numbers can be wrong a bit, it was hell of time ago) for mp3enc 128Kbit encoding.
But except some strange artifacts mp3enc produce here and threre, it is close to perfect tongue.gif (about artifacts - see ff123 page)
Still, it really can loose to last Lame (3.96), this version showed spectacular results (really unawaited) in the last Roberto test
So it seems we can congratulate Gabriel and other Lame team members - at the end Lame surpass FhG in quality ! rolleyes.gif
It will be interesting to see your test results...
Go to the top of the page
+Quote Post
plonk420
post Jul 18 2004, 10:06
Post #10





Group: Members
Posts: 312
Joined: 6-October 01
Member No.: 227



speaking of which, i tried playing with L3Enc, but it would just crash no batter what i tried...
Go to the top of the page
+Quote Post
westgroveg
post Jul 21 2004, 08:31
Post #11





Group: Members
Posts: 1236
Joined: 5-October 01
Member No.: 220



CODE
ABC/HR Version 1.0, 6 May 2004
Testname: FhG Family BlackBird

1L = C:\Test_Samples\BlackBird\CEP 2.0 BlackBird.wav
2R = C:\Test_Samples\BlackBird\Fastenc 1.02.wav
3L = C:\Test_Samples\BlackBird\Fastenc 1.02 -hq.wav
4L = C:\Test_Samples\BlackBird\MP3Enc 3.1.wav

---------------------------------------
General Comments:
Sample 1=  I think I can hear a very slight click but the max bandwidth is too low making it dull sounding is what I can ABX, Easy to ABX

Sample 2=  Can't hear a click, slightly less dull than sample 1, Easy to ABX

Sample 3= Not dull sounding but click makes it easy to ABX, I prefer this to the dull sound, Very easy to ABX though

Sample 4=More Dull sound than 3 + click is present as in sample 3

1 & 2 are similar 3 & 4 are similar, I prefer Sample 3.
---------------------------------------
ABX Results:
Original vs C:\Test_Samples\BlackBird\CEP 2.0 BlackBird.wav
   16 out of 16, pval < 0.001
Original vs C:\Test_Samples\BlackBird\Fastenc 1.02.wav
   16 out of 16, pval < 0.001
Original vs C:\Test_Samples\BlackBird\Fastenc 1.02 -hq.wav
   16 out of 16, pval < 0.001
Original vs C:\Test_Samples\BlackBird\MP3Enc 3.1.wav
   16 out of 16, pval < 0.001
C:\Test_Samples\BlackBird\CEP 2.0 BlackBird.wav vs C:\Test_Samples\BlackBird\Fastenc 1.02.wav
   8 out of 10, pval = 0.055
C:\Test_Samples\BlackBird\Fastenc 1.02 -hq.wav vs C:\Test_Samples\BlackBird\MP3Enc 3.1.wav
   10 out of 10, pval < 0.001

Well I belive this is fastenc -hq's worst problem sample & it really is not that bad to my ears, I have tested on some other samples too & I have to say I prefer fastenc -hq more than the others FhG encoders maybe because it has more fidelity, I'm not sure. I know I'm sensitive to high frequencies & that's where it shines also a higher max bandwidth doesn't help the other encoders much I'm not sure why. It may already be common knowledge as the NT crash bug that I was not aware of but in 2 samples I have tested so far fastenc's stereo image has collapsed where fastenc -hq does not, it's possible it doesn't have this bug. My vote would be for fastenc 1.02 -hq as the FhG encoder in the next MP3@128kbps public listening test.

This post has been edited by westgroveg: Jul 21 2004, 08:36
Go to the top of the page
+Quote Post
ff123
post Jul 21 2004, 16:46
Post #12


ABC/HR developer, ff123.net admin


Group: Developer (Donating)
Posts: 1396
Joined: 24-September 01
Member No.: 12



QUOTE (westgroveg @ Jul 20 2004, 11:31 PM)
Well I belive this is fastenc -hq's worst problem sample & it really is not that bad to my ears, I have tested on some other samples too & I have to say I prefer fastenc -hq more than the others FhG encoders maybe because it has more fidelity, I'm not sure. I know I'm sensitive to high frequencies & that's where it shines also a higher max bandwidth doesn't help the other encoders much I'm not sure why. It may already be common knowledge as the NT crash bug that  I was not aware of but in 2 samples I have tested so far fastenc's stereo image has collapsed where fastenc -hq does not, it's possible it doesn't have this bug. My vote would be for fastenc 1.02 -hq as the FhG encoder in the next MP3@128kbps public listening test.
*


fastencc without -hq does have the stereo collapse bug, as you found out. It's not recommended in this mode. fastencc -hq is liked by many with good high frequency hearing, but it is verrry slow. And you will get occasional glitches, knocking sounds, blips, etc., which may or may not disturb you too much, depending on your hearing and preferences. I would also like to have seen it compared against other top-knotch mp3 encoders.

ff123
Go to the top of the page
+Quote Post

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: 18th September 2014 - 22:05