Hello. There was some discussion about LAME VBR encoding at Un4seen forums. A user seems to be experiencing cutting off of high frequencies using the VBR mode. I haven't done much encoding for a while myself, and am not an expert otherwise either, so I thought someone here would surely have an idea of what's going on .

Here's a direct link to the thread http://www.un4seen.com/forum/?topic=8588.0

Pay close attention to the posting by Pike84, then pretty much ignore anything else that was said.



Edit: Oops - I guess you are Pike84.

Hehe.. Yeah, I'm the one

It is typical for Lame to cutoff high frequencies when encoding in CBR, ABR, and VBR. I am sure that there are some people over there on un4seen who are looking at their music rather than listening to it. Simply tell them that we don't listen to music by looking at the frequency plots, we listen to our music by using our ears. It doesn't matter how good or bad a lossy encoder can replicate the frequencies, all that matters is how the files sound. I can produce a good looking 320kbps CBR Lame mp3 file but it will sound terrible simply because the encoder is using frequencies that are too high. It will look good in the frequency plot though but it won't sound good.

When in doubt, ABX.

I almost feel like joining that forum for the sole purpose of giving gigalot a good verbal kicking. He really does talk total $#!te.

Cheers, Slipstreem.

Isn't that what the Y switch does and from looking at the HA wiki, the Y switch is turned on by default at V3 to V9. Maybe thats what the person is noticing.

Oh please dont turn this thread into another high-frequency cut-off debacle. There are so many of these on the board already, I'd suggest Pike rather read those instead.

If you have a few good audio CD which codec and setting you try to store music collection for yourself?

Gigalot: Welcome to the board. Having read your posts at unseen I can assure you that you need to contain yourself quite a bit more here. I am certain you read the TOS before joining, so I wouldnt assume it should be any problems.

Either way, you make a valid point, its always best to test for yourself.

@gigalot: Do yourself and everybody else a favour and read the Hydrogenaudio LAME WIKI because you don't seem to have the faintest idea what you're talking about when it comes to the LAME MP3 encoder.

Cheers, Slipstreem.

You have found the answer yourself: Lame (I suggest to use latest version 3.98 which has various improvements over 3.97) -V0 seems to be the most appropriate Lame setting for you.
As for the high frequency debate: With a lowpass that starts way beyond 16 kHz and with no -Y option active it's hard to believe you will hear (abx!) a HF problem. Don't care about very young people with exceptional HF hearing unless you belong to this group (and don't think you belong to this group unless you can abx it).

My personal choice for encoding a precious CD: lossyWAV --extreme preprocessing before usimg FLAC. It's lossyWAV --standard for less precious tracks (the standard case as it says), or lossyWAV --portable or Lame 3.98 -V0 (depending a bit on my actual state of mind) for music I want to keep but which I don't consider precious. It's all on the background that I don't keep a lossless archive any more, so my quality demands are pretty high.

Thanks a lot kick my uss! I`l go to read more about Lame and then try to download newest mp3sencoder which is available now. See you soon, king regards, Gigalot

Maybe you can participate in my upcoming MP3 listening test at 128 kbps.

@sizetwo:

As I said, I haven't encoded mp3s in a long time myself, I'm not the one having problems - I was just doing a favour for gigalot .

I know folks here are tired of poorly based arguments about sound quality and the like, but please try to avoid flaming - everyone's been new to these things at one point in their life .

Can we please please see some ABX tests before anymore assumptions are made about any settings. This is especially aimed at gigalot. We've all experienced the affects of expectation bias and I think you'll be surprised at it's effects once you've learned how to conduct an ABX test.

Foobar2000 has a built in ABX comparator. Queue up your two files, highlight them both, right click, Utils > ABX two tracks.

Cheers.

I could not agree more ! I was very surprised that the -k switch did not kick in yet

Oops! I need to reinstall Foobar to switch on ABX comparator

These days it's pretty hard to ABX the latest lossy encoders (aac, mp3-lame, fhr-mp3, itunes, ogg). Hasn't someone said in here that they were thinking to drop the 128kbps tests and move on to 96kbps, for the simple fact that it was unbearable to conduct the tests at 128kbps because of the improvement level the lossy encoders reached? With the exception of the well known killer samples, which is good to remind that you won't be listening to killer samples in 99.999% of the time you listen to music, you will be fine. Unless you are encoding your music below 128kbps, you should not worry about any modern codec, let alone if you are using insane presets like V0 and 320kbps.

Multiformat at 128 kbps probably doesn't make sense unless we find some hard samples (not killer samples for specific codecs / encoders, but general difficult to encode material).

I found no differences between Lame and mp3sencoder at ~199 vbr settings and at 128 cbr too durind foobar ABX. Both codecs seems pretty good and useful. I only suspect abut differences because the spectrums is not the same. May be only special passages or white noise can show some differences or if one track is encoded-decoded-encoded many times to encrease differences but not shure.

The next is i have no experiance in ABX.

As for me, i was surprised when can`t found any differences at pop musical track between 256 cbr Lame and ~65 kbit/sec MP4 SBR (Nero)! I think if you hear any differences it is realy BIG differences ~20-30% minimum. Not sure newest codecs can have it so much! I found only 64kbps MP3pro "eats" some sounds(instruments) and has realy big differences.

My English isn`t native and poor, excuse me if something is not correct!

I don't think that 128kbps MP3 is an universal solution for everybody (and every piece of equipment or music genre).
Listen to anything which has real hihat/cymbols/acoustic guitar in it and 128 kbps MP3 is not going to suffice.
http://martel.ic.cz/bordel/clavicula_sample.flac
If not anything else the 128kbps MP3 lowpass itself makes quite a difference... Your mileage may vary depending on your equipment/ears, though.

Are you sure that the MP3pro was properly decoded, including the SBR? I'm not saying that wasn't, only that this is a common mistake.

I use in_mp3pro.dll plugin v1.2 in XMpayer & 1by1. Great differences detectet vs Foobar2000`s common mp3. Realy poor sound if no mp3pro

Low cost sound cards often has distortions on very high frequences wich can produse realy audible artifacts in sound if it`s bitrate is high. At such situation 128 mp3 with lowpass 1600 seems better than 320. But it depends of amplfier or sound card quality, not Lame or Fraunhofer encoder.

Ahem... XMPlay

I took clavicula_sample.flac made clavicula_sample.mp3 128 amd clavicula_sample.mp4 ~64 SBR.
Then, converted all to 3 WAV. I can hear some wery small differences but realy don`t know which one is better if random mixed. May be i need to change my headphones or ears Or very hot and noisy toay for audio tests

Xmpay is good for MP3pro, no needed to switch between mp3 & mp3pro plugins as winamp if you don`t like mp3pro decoder always "ON" for all mp3

For Pike84 only: I mad skin modification for Xmplay by myself if you XMP fun. What you think about? http://rapidshare.com/files/132295755/WMP1..._Black_.7z.html Black skins always has better sound vs colored eyes can beats ears if bad skin

Have you tried performing an ABX test with various LAME VBR -V settings yet? It would be interesting to know just how far you can go down the scale before hearing any difference with various test samples of your own choosing.

Cheers, Slipstreem.

Well, as far as i remember, the purpose of MP3pro wasn't transparency (as with most other lossy codecs at that bitrate) but instead making differences as minor and "non-annoying" as possible. So, if you can hear slight differences, but cannot decide, which one you "like better", then MP3pro fulfilled its purpose.

Thankyou for taking the time to ABX and reporting your findings so far. Your english isn't so bad and it's making the effort that counts!

@gigalot:
If you want to learn something about mp3 artifacts (encoding errors), I can still highly recommend
ff123's Artifact Training Page
It shows you some common mp3 artifacts in difficult to encode samples.
Although it's a few years old and the encoder used is not the newest, the page was very useful to me.

edit:
I just checked the above and found that some links to the audio files are not functional anymore.

If you want to learn something about mp3 artifacts (encoding errors), I can still highly recommend
ff123's Artifact Training Page
It shows you some common mp3 artifacts in difficult to encode samples.
Although it's a few years old and the encoder used is not the newest, the page was very useful to me.

Thank you! I`ll try all tests. It was easy to determine which one is "castanets insane" vs 128k/bit new Fraunhofer mp3sencoder:

foo_abx 1.3.3 report
foobar2000 v0.9.5.4
2008/07/28 21:26:00

File A: F:\Documents and Settings\Alex\Desktop\castanets_insane.mp3
File B: F:\Documents and Settings\Alex\Desktop\castanets.mp3

21:26:00 : Test started.
21:41:05 : 01/01 50.0%
21:41:24 : 02/02 25.0%
21:41:50 : 03/03 12.5%
21:42:14 : 04/04 6.3%
21:42:42 : 05/05 3.1%
21:42:54 : Test finished.

----------
Total: 5/5 (3.1%)

and if you try "bibilolo" sample from Nero developers (thk Menno!) you can hear "Birds" trills on 128k/bit but somethins like "insects" sound on mp3pro 64k/bit. It realy "eats" middle-high frequences, or converts to noise.

foobar2000 v0.9.5.4
2008/08/01 00:29:57

File A: I:\sample3\sample3\castanets_m4.mp3
File B: I:\sample3\sample3\castanets_V5.mp3

00:29:57 : Test started.
00:31:54 : 01/01 50.0%
00:32:15 : 02/02 25.0%
00:33:05 : 03/03 12.5%
00:33:26 : 04/04 6.3%
00:34:10 : 05/05 3.1%
00:34:35 : Test finished.

----------
Total: 5/5 (3.1%)

Fraunhofer is realy perfect on this sample! Lame has problems in silent fragments. I can hear "lowpass switches artifacs?" in silent fragments of sample. Like old Philips dynamic noise limitter DNL in low class tape recorders.
I suspect, Lame has such problem on high bitrates too.
samples is here: url=http://rapidshare.com/files/133890692/Castsnets.7z.html

Fraunhofer m4 is ~151 kbit/sec VBR, Lame V5 is ~155 kbit/sec VBR and has easy detected audible artifacts.
But it was another artifacts in compare with 128 kbit CBR which has a problem with castanets sound. Lame V5 has something i can hear as "modulation". Fraunhofer seems clear at ~151

I don't know that I would agree with that. I took a few CD's (AC/DC - Back In Black for example) and ABXd LAME V3, Nero AAC Q.45, and Itunes AAC 128kbps. I wanted to get a feel for my level of transparency without doing 10 or 15 different test, I chose those 3.

The Itunes AAC was easily identifiable against the original WAV in every case with numerous artifacts. The LAME MP3 and Nero AAC were nearly impossible to ABX (granted at higher bitrates). I don't have an expert ear by any stretch, but 128kbps hasn't reached the level of transparency just yet in my opinion. I didn't try LAME at 128 CBR so I can't comment on that.

Castanets WAV vs WMA V2 ~130 VBR 2 pass encoding:

foo_abx 1.3.3 report
foobar2000 v0.9.5.4
2008/08/06 03:32:35

File A: F:\Documents and Settings\Alex\Desktop\castanets.wav
File B: F:\Documents and Settings\Alex\Desktop\castanets.wma

03:32:35 : Test started.
03:33:45 : 01/01 50.0%
03:35:21 : 02/02 25.0%
03:36:08 : 02/03 50.0%
03:36:56 : 03/04 31.3%
03:38:04 : 04/05 18.8%
03:38:29 : 05/06 10.9%
03:39:29 : 05/07 22.7%
03:40:02 : 06/08 14.5%
03:41:30 : 07/09 9.0%
03:43:14 : 08/10 5.5%
03:43:25 : Test finished.

----------
Total: 8/10 (5.5%)

It is difficult to hear differences between them but possible

Castanets though, is THE pre-echo killersample.... it represents pretty much one of the worst pre-echo scenarios possible. So that should be taken into account, when relating such results to "everyday listening" - its a worst case scenario, not something "usual".

Yes, it`s true!
All ~130 WMA, ~130 OGG and ~151 MP3 VBR is pretty good for most people at most "real" audio sound tracks and is better than 128 cbr MP3. Many people likes 320kbit/sec only because they are sure about it`s stable quality.
Newest codecs is so good that ~150 VBR hasn`t audible artifacts for 99.9% guys!

Fraunhofer mp3sencoder CBR 256kbit/sec very easy beats Lame 3.98 ABR 160kbit/sec (Lame VBR at such bitrate is realy bad for me!) at "real" music track. It is women voice with soft music.

foo_abx 1.3.3 report
foobar2000 v0.9.5.5
2008/08/13 22:13:18

File A: F:\Documents and Settings\Alex\Desktop\Track10fr.mp3
File B: F:\Documents and Settings\Alex\Desktop\Track10.mp3

22:13:18 : Test started.
22:14:54 : 01/01 50.0%
22:16:17 : 02/02 25.0%
22:17:13 : 03/03 12.5%
22:18:38 : 04/04 6.3%
22:20:50 : 05/05 3.1%
22:20:58 : Test finished.

----------
Total: 5/5 (3.1%)

Lame is best encoder? I don`t think so....

Why don't you use Lame 3.98 ABR 256 for a fair comparison? As for VBR I'd also give -V0 a try.

If you test LAME 3.98 CBR 256 vs. LAME 3.98 ABR 160
You may well find that LAME 3.98 beats LAME 3.98.
What does that tell you?

C.

Ah! Halb27 - too quick!

Sorry but.... how is that even a fair comparison? CBR 256 vs ABR 160??
And what were you busy with, to take 7 minutes for 5 trials, to ABX between both, if the quality is " is realy bad for me!" (as you say)

I sense the smell of a troll ... A quite devious, ABX'ing troll ...

Calm down, everyone. Maybe gigalot will post a sample.

A sample of the p!$$ he's taking?

Surely it should be obvious to anybody that the same bitrates should be used when comparing two encoders of the same format. For goodness sake!

Cheers, Slipstreem.

Why i tryed this test guys?
Because i was sure that fraunhofer 256 kbit/sec and Lame 3.98 160 abr (which results real average bitrate ~ 145kbit/sec) has the same quality. Both files seems good but ABX show not the same. It was surprise for me!

What was the difference? Realy don`t know, but something... can ABX it! Not enought transparent? Vibration? Modulation? ??? Realy impossible to say what it was.
Try it for yourself if you suspect i do something wrong.
Lame 256 vs fraunhofer 256? I can`t ABX any differences (castanets). Lame 320 vs Lame V0? I can`t hear differences.

I can post the sample, of cource. It is big, but i can upload it to rapidshare.

Only upload 30 seconds or less, preferably in a lossless file.

It doesn't matter what bitrate you choose to use, just use the same bitrate with both encoders. I fail to understand your logic of deliberately strangling LAME and then whinging that it can't compete with another MP3 encoder running at a much higher bitrate.

Cheers, Slipstreem.

I suspect Slipstreem has a 6-th sense whem speaks about my logic!
I am from Tbilisi, a few days war was around us and it stoped at 13 aug. I was ful of adrenalin and it was a good opportunity to test that adrenaline exacerbates hearing.
But anyway, i was sure abr 160 and cbr 256 is the same. Wrong? But I hope that this test was useful.

I'm afraid the test wasn't useful, for the reasons that Slipstream, and others (me included) have pointed out.
I'm glad you don't have a say in the Olympics; to determine the best runner all partipants would be running different distances.

C.

Pure track from CD:

Moderation: Link removed.

frh sample was 256kbit cbr joint stereo and Lame sample was 160kbit abr (96kbit min / 320kbit max).
Upload these samples too?

AAAARRRRGGGGHHHH!

Cheers, Slipstreem.