Hello everone,

I am having some strange issues with lame. It all started when I noticed the intro of a song encoded with --alt-preset standard wasn't sounding quite right.

To compare I encoded it with CBR160 instead, and it sounded alright, so I figured the vbr algorithm wasn't allocating enough bits in that section of the song.
So I added the -b 160 flag to make sure that at least 160kbps where used.

Strangely enough, the CBR160 file still sounds more close to the original and I can't figure out why!

So if anyone has any idea please tell my, because I want to select the best setting whilst keeping the files small for my portable player. And I don't see how a vbr file that is encoded with 160-320kbps can sound worse that a 160 file unless there is something wrong in the vbr mode.

Here is the troubling section: The vbr file has an unpleasing "wet-vibrato" effect on the guitar, same as when encoding with low bitrate (albeit less).


uncompressed.wav
CBR160.mp3
-V 2 -b 160.mp3

The two mp3 files are the same I use Lame 3.96.1 with settings
--preset cbr 128 --scale 0.4 -h It is good are my listening wile going for a
job and does not skip. For anything else -V 2 or --preset standard is great.

I notice files with clipping in will somtimes wobble on portables.

double post error

correct link to the second mp3 file:
V2-b160.mp3

Edit:
I can't ABX this, at least not at the moment with the TV on behind me...
Btw: What is it we are listening to? Sounds familiar...

Fixed...the link, sorry about that

It's the intro of the song "Under Your Spell" from the soundtrack of the Buffy episode "Once More With Feeling". I've not once failed to ABX the vbr file from 160kbps/192kbps and 256kbps CBR files, or the original wav.

The -b 160 switch did nothing for the quality of the section. I also tried the latest lame compile (3.96.1) with the same results.

Edit:

I'm of course not ABX'ing on the portable player Using the MAD decoder for Winamp and Sennheiser HD 212 PRO headphones plugged into an Audigy 2.

Edit 2:

I also noticed that the original sounds more "open" than the compressed versions, the overall tonality is notably different.
Strange, when I compress it to a 135-215kbps wma, I can't distinguish it from the original.

3.90.3's APS is not as perfect as many people believe, but it is a well balanced high quality preset. Therefore there will certainly be songs that a simple mid-to-high bitrate CBR setting can handle better than APS, but those songs are more exceptions than rules. I have encountered such songs as well. But on average, APS is supposed to be better overall.

If you are interested in smaller file sizes, you might also want to try LAME 3.97a4 (or 3.96.1)'s VBR, and see if you like it. For portable use in noisy environments, I would start with simple "-V 7" and if it's not good enough, go up to -V 6, -V 5, -V 4, etc. Some people might suggest adding "--athaa-sensitivity 1".

What I don't understand is why CBR160 still sounds better than vbr even though I am enforcing a minimum of 160kbps by using the -b 160 parameter.

What I don't understand is why CBR160 still sounds better than vbr even though I am enforcing a minimum of 160kbps by using the -b 160 parameter.
[/quote]


Can you have a trial, if a Lame alt-preset extreme or insane (320 cbr) encode solve stepwise the problem, you detected with preset standard ?


I think, the reason, why adding -b160 to increase the min. bitrate as "safety" doesn't work with presets, is following:
there is a bitrate reservoir, and if the "preset" decided, that only certain bits are necessary, then the preset "keeps the additional bits in mind", and allocates them later, when the preset thinks, it can use them.
or even they aren't used additionally, they are used in next frames for the normal bits-allocations, so that you cannot influence the presets' bit allocations much.

So, you will notice, that the average bitrates don't increase noticeably, if you try to enforce the min. bitrate by -b160.

Yes that particular problem is solved at -V 1, but then the file becomes very large.



Makes sense, but the CBR routines also use the bit reservoir and CBR160 should be atleast as aggressive as -V 2.

In addition, I've not been able to determine why the overall "sound" of the mp3-files have changed. It sound like they've been through an equaliser and the stereo image appears narrower, while an wma encoded file sounds more like the original. Gonna try comparing a 320kbps file later and see if the issue persists.

Could you please provide ABX results?
I'm having a problem hearing the artifact you described.

I've just tried ABXing this track and couldn't detect any difference at all (at aps with 3.90.3), after 15 minutes of trying.

I'd like to see some ABCHR results for this one.

I got something on 3.96.1 aps.

pos 6.6-9.3 sec - listen when the string hits 'boom' - its smeared and not as sharp.


foo_abx v1.2 report
foobar2000 v0.8.3
2004/12/14 00:33:35

File A: file://H:\temp\abx tests\amnesia trans\uncompressed.wav
File B: file://H:\temp\abx tests\amnesia trans\uncompressed.mp3

00:33:37 : Test started.
00:34:42 : Trial reset.
00:37:08 : 00/01 100.0%
00:37:57 : 01/02 75.0%
00:38:49 : 02/03 50.0%
00:40:00 : 02/04 68.8%
00:41:06 : 03/05 50.0%
00:41:37 : 04/06 34.4%
00:42:00 : 05/07 22.7%
00:42:18 : 06/08 14.5%
00:42:51 : 07/09 9.0%
00:43:04 : 08/10 5.5%
00:43:13 : 09/11 3.3%
00:43:58 : 10/12 1.9%
00:44:33 : 11/13 1.1%
00:45:01 : 12/14 0.6%
00:45:17 : 12/15 1.8%
00:46:16 : 13/16 1.1%
00:47:09 : Test finished.

----------
Total: 13/16 (1.1%)