Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Lame 3.98a11 (Read 25640 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Lame 3.98a11

My results for 3.98a11:

-V2:
Birds: not annoying to me
harp40_1, trumpet: slightly annoying to me
herding_calls, eig, lead-voice: annoying to me

The lead-voice tremolo is still there.

-V1:
Birds: not annoying to me
harp40_1: very slightly to slightly annoying to me
trumpet: slightly annoying to me: kind of tremolo effect here too
herding_calls: slightly annoying to annoying to me
eig, lead-voice: annoying to me

-V0:
Birds: not annoying to me
harp40_1: very slightly annoying to me
trumpet: slightly annoying to me (tremolo like effect)
herding_calls: slightly annoying to annoying to me
eig, lead-voice: annoying to me

--alt-preset 250 (125 in the case of lead-voice):
Birds, harp40_1: not annoying to me
lead-voice, herding_calls, trumpet: very slightly annoying to me (lead-voice tremolo audible here too at a very low level, trumpet not perfect as well though it's not exactly the kind of tremolo audible with -V0)
eig: slightly annoying to annoying to me


So the tremolo effect is audible with trumpet too though at a less exposed level.
lame3995o -Q1.7 --lowpass 17

Lame 3.98a11

Reply #1 –
I don't hear any tremolo with lead-voice -V2. Instead, I hear a sort of roughness (sort of sounds like pre-echo).
Code: [Select]
foo_abx v1.2 report
foobar2000 v0.8.3
2006/12/29 22:05:45

File A: file://F:\Downloaded Files\lead_voice.flac
File B: file://C:\lead_voice.mp3

22:05:47 : Test started.
22:06:42 : 01/01  50.0%
22:06:52 : 02/02  25.0%
22:06:59 : 03/03  12.5%
22:07:09 : 04/04  6.3%
22:07:19 : 05/05  3.1%
22:07:28 : 06/06  1.6%
22:07:47 : 07/07  0.8%
22:07:53 : 08/08  0.4%
22:08:08 : 09/09  0.2%
22:08:13 : 10/10  0.1%
22:08:30 : Test finished.

 ---------- 
Total: 10/10 (0.1%)


Effect is still present on -V1 and -V0, 10/10 on both tests. It didn't seem to get any better. Failed to ABX --alt-preset 250.

eig, highly annoying on -V2, -V1, and -V0, 10/10 on all tests. I didn't post logs for V2, V1 or V0 because it was just unnecessary for this sample. If you want, I can do it anyway. --alt-preset 250 actually is much less annoying, though still ABXable.

Code: [Select]
foo_abx v1.2 report
foobar2000 v0.8.3
2006/12/29 22:26:35

File A: file://F:\Downloaded Files\eig.wv
File B: file://C:\eig.250.mp3

22:26:38 : Test started.
22:26:55 : 01/01  50.0%
22:26:59 : 02/02  25.0%
22:27:03 : 03/03  12.5%
22:27:09 : 04/04  6.3%
22:27:13 : 05/05  3.1%
22:27:17 : 06/06  1.6%
22:27:21 : 07/07  0.8%
22:27:26 : 08/08  0.4%
22:27:31 : 09/09  0.2%
22:27:37 : 10/10  0.1%
22:27:38 : Test finished.

 ---------- 
Total: 10/10 (0.1%)


250 vs v0.
Code: [Select]
foo_abx v1.2 report
foobar2000 v0.8.3
2006/12/29 22:16:56

File A: file://C:\eig.250.mp3
File B: file://C:\eig.v0.mp3

22:16:59 : Test started.
22:17:15 : 01/01  50.0%
22:17:17 : 02/02  25.0%
22:17:20 : 03/03  12.5%
22:17:22 : 04/04  6.3%
22:17:25 : 05/05  3.1%
22:17:29 : 06/06  1.6%
22:17:31 : 07/07  0.8%
22:17:33 : 08/08  0.4%
22:17:36 : 09/09  0.2%
22:17:38 : 10/10  0.1%
22:17:39 : Test finished.

 ---------- 
Total: 10/10 (0.1%)


harp40_1:
-V2, I had a hard time ABXing this.
Code: [Select]
foo_abx v1.2 report
foobar2000 v0.8.3
2006/12/29 22:18:41

File A: file://F:\Downloaded Files\harp40_1.wav
File B: file://C:\harp40_1.v2.mp3

22:18:44 : Test started.
22:19:15 : 01/01  50.0%
22:19:19 : 02/02  25.0%
22:19:23 : 03/03  12.5%
22:19:27 : 04/04  6.3%
22:19:31 : 04/05  18.8%
22:19:34 : 05/06  10.9%
22:19:39 : 06/07  6.3%
22:19:43 : 07/08  3.5%
22:20:00 : 08/09  2.0%
22:20:16 : 09/10  1.1%
22:20:18 : Test finished.

 ---------- 
Total: 9/10 (1.1%)


I failed to ABX -V1 (15/20), so I'm not even gonna try V0. I don't like the sound of harpsichords, so this is a sample I won't be testing again.

I can't find Birds, herding_calls, or trumpets, can you direct me to the thread with them?

I could also try normal music with a11, but I'm sure to fail the ABX on them.

Lame 3.98a11

Reply #2 –
Changes in LAME MP3 Encoder 3.98 alpha 11:

• Alexander Leidinger:
- Add TLEN (ID3v2) support (Submitted by: Linus Walleij).
- Add number of total tracks per album (ID3v2) support (Submitted by: Kyle VanderBeek).
- Some seatbelts for overflowing arrays in the ID3v2 support.
- Update the RPM spec (Submitted by: Kyle VanderBeek).
- Fix some mem-leaks in the error case.
- Update to newer autotools versions.
- Update to use a recent libsndfile (submitted by libsndfile author).
- Intrinsics support enabled for gcc

• Robert Hegemann:
- Fixed some console printing problems
- Fixed: in case of not enough bits the new vbr code incorrectly used old vbr routine
- Fix regarding short blocks ATH values
- Improved ATH adjustment in low volume cases

• John33:
- Fixed mp2 and mp3 decoding: For mp3 and mp2 decoding, this now yields the same output as foobar2000 but the error checking remains unchanged

• Gabriel Bouvigne:
- VC8 project files
- Added support for x64 under VC8
- Restricted MPEG 2.5 to 64kbps frames

• Takehiro Tominaga:
- SSE version of FFT

Lame 3.98a11

Reply #3 –
Quote from: Firon on
I can't find Birds, herding_calls, or trumpets, can you direct me to the thread with them?


I found them here: http://www.hydrogenaudio.org/forums/index....showtopic=39314

Birds: http://www.hydrogenaudio.org/forums/index....showtopic=39335
herding_calls: http://www.hydrogenaudio.org/forums/index....showtopic=37002
trumpets: http://www.hydrogenaudio.org/forums/index....showtopic=39334

Why is there no sticky with all (or most important) test samples yet... (i.e. with links to the threads in Uploads)?

Lame 3.98a11

Reply #4 –
Quote from: Fandango on
...Why is there no sticky with all (or most important) test samples yet... (i.e. with links to the threads in Uploads)? ...

The problem is: what are important test samples? Cause it depends on the context.

For testing 3.98 those problems are especially relevant which are known that 3.97 yields worse results than can be expected.
Prominent samples in this respect are
- Wombat's samples like Birds (3.98 has improved upon it perfectly IMO since 3.98a3)
- herding_calls (which IMO belongs to the same class of problems, but 3.98 should still improve on this sample)
- trumpet (which according to guruboolez belongs to another problem class not that important; 3.98 behaves pretty well already with it but I guess it can still be improved)
- lead-voice  (which IMO currently is performing worst with 3.98 among the samples tested).

harp40_1 and eig aren't samples tailored for 3.98. As they are difficult samples it's interesting anyway how 3.98 performs, and it does so pretty well. Some day for these samples I compared 3.98 alpha (around a8) with 3.97 and 3.90.3, and eig was best with 3.98.

Apart from samples like harp40_1 and eig where temporal resolution is the problem there are hardly any samples which are really universally problematic to different encoders. Of course there are always samples which show up specific weaknesses of specific encoders. And there's a grey zone: herding_calls and trumpets are not really universally problematic, but it's remarkable that many encoders of various formats require a higher quality setting to these simple-sounding samples than is usually required.
lame3995o -Q1.7 --lowpass 17

Lame 3.98a11

Reply #5 –
Quote
This Account Has Been Suspended
Please contact the billing/support department as soon as possible.

Lame 3.98a11

Reply #9 –
Quote
Added FFTSSE and FFT3DNOW assembler code from Lame4 branch


A little bit off topic, can somebody enlighten me how to trigger SSE asm code with MinGW. I've tried with
Code: [Select]
$ ./configure --enable-nasm CFLAGS="-march=pentium3 -mmmx -msse -O3 -fomit-frame-pointer -pipe"

but the .exe told me it uses mmx asm and sse (not sse asm).

Lame 3.98a11

Reply #10 –
Comparison between LAME 3.97 and 3.98a11 based on 40 samples encoded with -V5 --vbr-new:
>>here<<


All 40 samples (60 MB) are temporary uploaded here.

Lame 3.98a11

Reply #11 –
Quote from: guruboolez on
Comparison between LAME 3.97 and 3.98a11 based on 40 samples encoded with -V5 --vbr-new:


the presentations of your test-results are getting better and better 
nice reading btw

Lame 3.98a11

Reply #12 –
This is really really interesting... Guru finds the latest alpha clearly outperforming 3.97 on problem samples but doing worse on classical music. This proves (or at least indicates) that one can not put too much emphasis on problem samples. (Even though Guru stated that you can't really draw any conlcusions) Real music is different. I hope the devs find this test most useful.
//From the barren lands of the Northsmen

Lame 3.98a11

Reply #13 –
It's also a strong hint on different behavior within different bitrate classes.
So for the 130 kbps average bitrate class 3.97 is superior with regular music - at least to fans of classical music.
For the very high average bitrate class (200 kbps+) it is expected that 3.98 behaves better as problematic cases are treated better and more regular music is not expected to behave badly in this range. This isn't proved however.
I guess 3.98 will still be improved - especially with regard to strange tremolo behavior.
lame3995o -Q1.7 --lowpass 17

Lame 3.98a11

Reply #14 –
Very interesting read

Gives me some ideas for my future ABX test.

Lame 3.98a11

Reply #15 –
As for guruboolez' test I'm quite happy to see there's more evidence that ABR can lower or even get away with problems existing when using VBR. This is the very reason why I prefer ABR so far when using Lame with very high bitrate.

ADDED:
But this is not a suggestion for prefering abr at ~130 kbps: I'm just doing a listening test (will report on it), but so far it looks like 3.98a11 abr132 is inferior to -V5 in a general view.
lame3995o -Q1.7 --lowpass 17

Lame 3.98a11

Reply #16 –
I'm just wondering if the strict ISO compatibility is one of the main reason that can cause regression of NORMAL (classical) music? From the ChangeLog, it's the only change that may have negative affect of quality.

Lame 3.98a11

Reply #17 –
At -V5 chances for this being the reason are pretty low.

As for the regression with classical music I think everybody has to judge for his own as everbody has a different experience, different preferences, and different hearing abilities.

I'm just doing a listening test with part of guruboolez' samples apart from mine and will report on it. I took (among others) the worst performing classical samples, and can abx most of the encoded results as being not transparent (with more or less all the encoders and settings I tried - I didn't only use 3.98a11 and 3.97). However to me differences against the original are all pretty subtle. I'd call them very slightly annoying and don't beleive I would hear them in a non-ABXing real world listening environment. I'm not emotionally concerned cause quality is good-enough to me in the 130 kbps bitrate range where I am prepared to accept compromises as long as they don't really prevent me from enjoying the music.

Sure this is only a statement valid for myself based on the background that I'm not used very much to listen to classical music and that I certainly don't have the golden ears. But this exactly what I mean: everybody has to find out for himself about the relevance of published listening tests.

I know from experience it is possible to increase the sensitivity for problems dramatically within 2 days. But I keep well in mind what Gambit wrote one day: he doesn't want to learn to differentiate subtle deviations from the original but wants to enjoy the music.

As for that to me pretty obvious (with ABXing) deviations from the original which I am afraid to hear in real world listening situations are most of concern.
This applies to
  • limited temporal resolution easily audible as in harpsichord samples or pre-echo samples like eig.
    It's a general transcoder problem and is especially error-prone for the mp3 format. However different mp3 encoders provide a significantly different quality.
  • the kind of errors audible among guruboolez' samples in Moon. This error is Lame specific. 3.97 behaves extremely badly, 3.98a3+ has improved a lot but is not really good yet when looking at herding_calls.
  • tremolo-like artifacts. I ran upon them when testing 3.98alphas at high bitrate settings, and in fact latest alphas are specifically error-prone to this. With my current listening test @ ~130 kbps however I found tremolo behavior is a bit more common also to other encoders than I imagined. Trumpet samples seem to be specifically affected. It's like that with 3.97, but also with 3.90abr, as well as with Helix' -V55 default and level's setting. Sure it makes a big difference to what degree the tremolo artifact is audible, and there are huge quality differences between the encoders.
lame3995o -Q1.7 --lowpass 17

Lame 3.98a11

Reply #18 –
I also have same experience as guruboolez's. I did short test of LAME VBR and FhG cbr128 with orchestra sample.
This could be the problem sample for LAME 3.97, but 3.98a makes thing worse.

Code: [Select]
ABC/HR for Java, Version 0.52b, 04 1・2007
Testname: Mahler - Symphonie 3 fin

Tester: haregoo

1L = E:\Desktop\128k test\Mahler - Symphonie 3 fin_LAME397_V3.mp3
2R = E:\Desktop\128k test\Mahler - Symphonie 3 fin_LAME398a11_V5.mp3
3R = E:\Desktop\128k test\Mahler - Symphonie 3 fin_FhG.mp3
4R = E:\Desktop\128k test\Mahler - Symphonie 3 fin_LAME397_V2.mp3
5L = E:\Desktop\128k test\Mahler - Symphonie 3 fin_LAME397_V4.mp3
6R = E:\Desktop\128k test\Mahler - Symphonie 3 fin_LAME397_V5.mp3

Ratings on a scale from 1.0 to 5.0

---------------------------------------
General Comments: 
---------------------------------------
1L File: E:\Desktop\128k test\Mahler - Symphonie 3 fin_LAME397_V3.mp3
1L Rating: 4.3
1L Comment: 
---------------------------------------
2R File: E:\Desktop\128k test\Mahler - Symphonie 3 fin_LAME398a11_V5.mp3
2R Rating: 2.7
2R Comment: severe distortion
---------------------------------------
3R File: E:\Desktop\128k test\Mahler - Symphonie 3 fin_FhG.mp3
3R Rating: 3.5
3R Comment: 
---------------------------------------
4R File: E:\Desktop\128k test\Mahler - Symphonie 3 fin_LAME397_V2.mp3
4R Rating: 4.1
4R Comment: 
---------------------------------------
5L File: E:\Desktop\128k test\Mahler - Symphonie 3 fin_LAME397_V4.mp3
5L Rating: 3.8
5L Comment: 
---------------------------------------
6R File: E:\Desktop\128k test\Mahler - Symphonie 3 fin_LAME397_V5.mp3
6R Rating: 3.0
6R Comment: 
---------------------------------------

ABX Results:
Original vs E:\Desktop\128k test\Mahler - Symphonie 3 fin_FhG.mp3
    10 out of 12, pval = 0.019


---- Detailed ABX results ----
Original vs E:\Desktop\128k test\Mahler - Symphonie 3 fin_FhG.mp3
Playback Range: 00.000 to 30.000
    6:54:12 PM p 1/1 pval = 0.5
    6:54:17 PM p 2/2 pval = 0.25
    6:54:29 PM p 3/3 pval = 0.125
    6:54:45 PM f 3/4 pval = 0.312
    6:54:47 PM p 4/5 pval = 0.187
    6:54:50 PM p 5/6 pval = 0.109
    6:55:02 PM p 6/7 pval = 0.062
    6:56:28 PM f 6/8 pval = 0.144
    6:56:31 PM p 7/9 pval = 0.089
    6:56:38 PM p 8/10 pval = 0.054
    6:56:42 PM p 9/11 pval = 0.032
    6:56:47 PM p 10/12 pval = 0.019

I'm a bit cofused about lossy of choice.

Lame 3.98a11

Reply #19 –
Quote from: haregoo on
I also have same experience as guruboolez's. I did short test of LAME VBR and FhG cbr128 with orchestra sample.
This could be the problem sample for LAME 3.97, but 3.98a makes thing worse. ...

According to your results 3.98a11 isn't a lot worse than 3.97, and it looks like you're more concerned about the better behavior of FhG.

First of all IMO your results show that in order to get at a very good quality it is important to use a higher quality setting - 3.97 -V3 was pretty good.
I think we can take this as a rule: use a higher quality setting (go into the average 160+ kbps range) and quality usually is so high with any good mp3 encoder that differences between encoders usually become negligible.
What is still of concern then is real erraneous behavior like 3.97's sandpaper problem (Moon, TomsDiner, herding_calls, ...) and current 3.98alpha's tremolo problem.

As for ~128 kbps: as you have a good hearing towards classical music: would you mind trying Helix -V55 with your sample (or -V60 cause it's hard to say whether
-V55 or -V60 matches better an average of ~130 kbps)? To me the result is fine - but this doesn't say much as I'm not much used to music like that and also have restricted hearing abilities (however I do know that Helix performs significantly better than any other mp3 encoder I tried on samples with potential problems with temporal resolution).
lame3995o -Q1.7 --lowpass 17

Lame 3.98a11

Reply #20 –
Quote from: Enig123 on
I'm just wondering if the strict ISO compatibility is one of the main reason that can cause regression of NORMAL (classical) music? From the ChangeLog, it's the only change that may have negative affect of quality.

I'm just wondering why is the strict iso switch forced. AFAIK there was no clear answer to this, is it because of some legal issues? Or is it because like 0.01% mp3 decoders had problems with the bit reservoir? If so, I see no reason why it should be forced (it could be on by default and you could deactivate it if you want).

Lame 3.98a11

Reply #21 –
Quote from: jmartis on
I'm just wondering why is the strict iso switch forced. AFAIK there was no clear answer to this, is it because of some legal issues? Or is it because like 0.01% mp3 decoders had problems with the bit reservoir?

It's because otherwise there is a decoding issue (ie no sound at all) with some versions of the FhG ACM decoding filters. Those versions are shipped with some windows version, so it's clearly a mainstream decoder.

Lame 3.98a11

Reply #22 –
I just finished my test on a selection of guruboolez' samples and added Birds, harp40_1 and herding_calls.

The most astonishing results I didn't expect:
  • Tremolo problems I attributed to 3.98 so far can be a rather universal issue.
  • The sandpaper noise like problem exists not only for 3.97.
For the details:

I hope nobody expects this to be an alternative test to guruboolez' cause of course guruboolez can differentiate the deviations from the original a lot better than I can.

I have different targets in mind:
  • I want to see to what extent the results of another person's listening test are relevant to me. I have a feeling since quite a while that it's pretty hard to judge about the meaning of the outcome of a listening test.
  • As I'm an ABR fan in the very high bitrate range I wanted to learn whether or not 3.98abr is an alternative to 3.98 -V5.
  • As so far to me 3.90 is the more robust alternative to 3.97/3.98 in the very high bitrate range I wanted to learn whether or not it is an alternative in the 130 kbps range too. As I dislike Lame VBR from any version before 3.98 I wanted to use 3.90.3abr.
  • Last not least I wanted to learn about Helix' behavior in the 130 kbps average bitrate range as this is my favorite mp3 encoder at high bitrate so far. Moreover I wanted to learn whether the default setting or level's setting should be preferred.
  • After having read haregoo's post I wanted t learn a bit about the FhG surround encoder. I did this after ending my test and tried it only on some of the samples where other encoders had remarkable problems with. So the FhG encoder did not really contribute in the test, I just wanted to learn something about it.
For the abr methods I used --alt-preset 132 as Lame 3.97/98 -V5 provides an average bitrate which IMO is slightly above 128 kbps. I don't want to make a science out of this, just consider abr 132 to be roughly equivalent.
With Helix it's harder to find a fair equivalent setting. From the Helix thread I learnt -V50 is a bit too low, and -V75 is suggested as yielding ~128 kbps. But according to my trials this is way too high, and I feel -V55 or -V60 is more appropriate. So as Helix default setting I chose -V55, and for level's setting I chose -V55 -X2 -SBT450 -TX0.
With 3.97 I used --vbr-new.

As a selection of Guruboolez' samples I chose those which came out pretty bad in his test (and which are not of an a priori restricted relevance like the gain boosted low volume samples):
1.04,  1.05,  1.06,  1.11, 1.13, 2.02, 2.03, 2.04, 2.08, 2.09, 2.11, 2.13, 1.3, 2.1, 2.2, 3.2, 4.1, 4.2 6.1, 6.2, and added Birds, harp40_1 and herding_calls to the problem sample class.

As for the listening results I want only report on the essential things, and I partially did it already in other posts.
As for the classical and various genre samples except for specific samples I will report about below to me things are pretty fine, that is adequate for the bitrate range. While I can abx a lot of encodings as being not transparent, things usually aren't severe to me. If I had to make a quality ordered list I'm not sure whether this list would be the same if I repeated my test one week later. Deviations from the original are usually so low to me that if I would notice them at all while real life listening without access to the original I don't think I would even call it 'slightly annoying'.

The most remarkable exception to this in the classical samples class is 1.06 solo viola de gamba. With this sample I hear a tremolo artifact with any encoder. At least in the last 2 seconds 3.97 isn't better to me than 3.98a11, 3.98a11abr is slightly better, best are 3.90.3abr, both Helix encodings, and FhG surround. When I first heard the tremolo effect on lead-voice with a 3.98alpha version I had  never imagined that such a kind of tremolo effect can be an issue to many encoders with the same sample.

The most remarkable sample of the mixed sample class is Tom's Diner. I'm not used so much to how special instruments are sounding, but deviations from a voice are easily noticed as we are all used to listen to human voices. 3.97's sandpaper noise problem effects voices, maybe that's what makes it an outstanding problem.
Other than that results of the different encoders in the mixed genre class are also pretty mixed to me with no other serious problem to me.

From the problem sample class it's remarkable that the two trumpet samples provide for a rather universal tremolo problem again. Now I know why I heard a tremolo effect on trumpet at very high bitrate which came as a surprise. Best are both Helix encodings and FhG which provide for a roughly acceptable quality.
4.1 tonal (BWV 1012 Allemande) has a tremolo issue too for 3.97, 3.98V5 and - at a lower degree - 3.98abr.
With eig both Helix encodings provided for the best quality. Same goes for 3.2 harpsichord.
Quite surprising was that the sandpaper noise problem in Moon isn't specific to 3.97, but is also a rather universal issue. It's clearly worst with 3.97, but can also easily be noticed with with 3.98 (abr worse than -V5), and 3.90abr. Best is Helix.
Birds has the sandpaper issue with 3.97, but also with 3.98abr and 3.90abr. 3.98V5, Helix and FhG are okay to me.
harp40_1 is best with Helix default and worst with 3.98abr.
Both Helix encodings are again best on herding_calls, and 3.98abr is worst.

3.98 ABR isn't better than 3.98 V5. On some samples it is but with some samples where it's worse it's very remarkably worse.

The tremolo problem isn't totally specific to the 3.98 current alphas though it's worst there.
As well isn't the sandpaper noise problem specific to 3.97 though it's definitely worst with 3.97.

The Helix encodings performed remarkably well. To me the default setting provides roughly the same overall quality as level's setting. I prefer the default setting a little bit. Among the encodings where I can hear differences between these version those which were in favor of the default setting were more definite to me: 2.04 experiencia and harp40_1.

Compared to 3.97/3.98V5 3.90abr is a good alternative judging from my test. However it's not competetive to Helix.

FhG surround encoder used with CBR128 didn't really participate in the test, but among the samples I tried belonged to the best encoders.

ADDED:

I should have done it before, but I checked out the equivalence of the settings of the different encoders a lot preciser right now.
As a test collection I took 125 full-length samples (26 samples of classical music, 29 samples of singer/songwriter music, and 70 tracks of pop music of various forms).
I checked the total file size cause that's what counts in the end when in the low/moderate bitrate range - differences in overhead with full-length samples aren't significant  anyway.

FhGsurround_CBR128: 501,0 MB
Lame3.97_V5vbrnew:  500,5 MB (a perfect match to CBR128)
Lame3.90_ABR132:    507,6 MB (roughly equivalent, but ABR130 would be more appropriate)
Helix_V55:                  490,8 MB (roughly equivalent)
Helix_V60:                  508,2 MB (roughly equivalent, a better match than V55).
lame3995o -Q1.7 --lowpass 17

Lame 3.98a11

Reply #23 –
Thanks for your interesting test Halb27! Wow, this really leaves everything hanging... Guruboolez has always favored LAME over Helix but this points in the opposite direction! I, personally, really like Helix because it's so fast and simple. Does this call for another test ~130 kbps among the mp3 encoders? I know people are gonna say "try it out for yourself..." But that answer won't cut it... Like I said, interesting...
//From the barren lands of the Northsmen

Lame 3.98a11

Reply #24 –
I hope we'll get a 128 kbps test in the near future.
In fact it was prepared already but was abandoned in favor of a 48 kbps listening test which attracted more interest.
robert is trying hard to get at 3.98 final before this test, and this test was also my motivation for comparing Helix default with level's setting.
This upcoming test is gonna be real interesting.
lame3995o -Q1.7 --lowpass 17