#384 low birate compression is better with 3.96

Quality
open
Libraries (86)
6
2011-03-05
2010-10-09
tiger man
No

Hi, i\'m compressing some wavs for a website at 64 kbps and i\'ve just discovered that from lame 3.97 and following compression is worst than 3.96.1
I\'ve attached two mp3, you can clearly hear it on the kick drum, with 3.96 is equal every shot while with 3.98 it changes overtime and it addsl also a little distortion sometimes.
Hope this helps, in the meantime i keep using 3.96.1.
command line is --cbr -q 0 -b 64
thanks for reading.

Discussion

  • tiger man

    tiger man - 2010-10-09

    compressed with 3.96.1

     
  • tiger man

    tiger man - 2010-10-09

    compressed with 3.98.4

     
  • tiger man

    tiger man - 2010-10-09
    • summary: low quality compression is better with 3.96 --> low birate compression is better with 3.96
     
  • Robert Hegemann

    Robert Hegemann - 2010-12-04
    • labels: --> Libraries
    • milestone: --> Quality
    • priority: 5 --> 6
     
  • Robert Hegemann

    Robert Hegemann - 2011-03-05

    Would it be possible to upload the corresponding uncompressed wave file?
    Else, as current CVS has some fix for some other CBR related report, could you try latest 3.99 alpha 14?

     
  • Robert Hegemann

    Robert Hegemann - 2011-03-05
    • assigned_to: nobody --> robert
     
  • tiger man

    tiger man - 2011-03-11

    Hi, here's the wav, can't upload here because of file size sorry:

    http://rapidshare.com/files/452074419/slice.wav

    I've tried with 3.99 alpha14, in this case it looks much better than previous versions but 3.96 is still better, on this file is barely noticeable but please try this second wav, is the same track but in another point:

    http://rapidshare.com/files/452074437/slice2.wav

    i've tried with 3.96, 3.98.4 and 3.99 alpha14 (all x86) using this command line

    lame.exe --cbr -q 0 -m s -b 80 *.wav

    3.96.1 - no noticeable issues
    3.98.4 - similar to 3.96 but it's less precise and sometimes it loses some high frequencies and gives an effect like a little saturation or "underwater"
    3.99 a14 - probably the worst one, it lose a lot of high frequencies compared to 3.96 and it seems the "underwater" effect is constantly active.

    i've tried compressing at 64kbps with 3.96 and it looks like it's better than 3.99a14 at 80kbps, it has less high frequencies but it does not have that ugly bubbly effect.

    maybe i'm missing something on the command line ?
    Hope this helps and thanks for getting into it

     
  • tiger man

    tiger man - 2011-07-20

    lame 3.96.1 --cbr -b 80 -q 0 -k

     
  • tiger man

    tiger man - 2011-07-20

    lame 3.98.3 --cbr -b 80 -q 0 --highpass 0.01 --lowpass 21.00

     
  • tiger man

    tiger man - 2011-07-20

    Hi, i have done more tests, with lame 3.97 and newer there are issues with cbr encodings, while vbr are ok.
    comparing this 2 command lines:

    with lame 3.96.1
    --cbr -b 80 -q 0 -k

    with any lame newer than 3.96.1 (currently i'm using 3.98.3)
    --cbr -b 80 -q 0 --highpass 0.01 --lowpass 21.00

    this is to make sure that no filters will be applied on both conversions, including resample.

    as you can hear from the new attached files, conversion with the old lame is quite good, very precise and almost all transients are reproduced fine, while on 3.98 it's quite noticeable the loss of precision and quality.

    the more you go up with kbps, the less this bad effect appears, unfortunately it looks it totally goes away only at 192kbps, it's very rare and barely noticeable at 160kbps.

    i've attached the 2 new tests, and for file size reasons a -V 0 mp3 that you can decode to wav in order to test by yourselves.

    thanks

     
  • tiger man

    tiger man - 2011-07-20

    i just want to add that this issue is not strictly related to this audio file, you can hear this issue on most electronic/looped music, expecially tracks with few sounds, where drums transients are clearly audible.

     
  • Riccardo Ferri

    Riccardo Ferri - 2012-03-17

    Hallo, i've noticed this issue also on 3.99.3, any news about it ?
    Also i have a request, would it be possible to include the older algorythm on newer lame releases, something like --cbr-old ? like you have done for vbr?
    It would be very nice as 3.96.1 is still the best encoder for cbr up to 160kbps, also the build i have was for ppc only so it doesn't works anymore under osx 10.7 Lion. Lastly, it's extremely slow on OSX (less than 1x with q 0).
    thanks.

     

Log in to post a comment.

Get latest updates about Open Source Projects, Conferences and News.

Sign up for the SourceForge newsletter:





No, thanks