Frequency Response Spectrum Analysis
for music encoded with LAME

Measurements are done using roundtrip WAV -> MP3 -> WAV audiofiles.
The graphs are no longer here, since they so do not apply to the recent LAME version.

VBR versus ABR and CBR :

Personally I don't like the thought behind ABR, I will never encode to MP3 with a target (average) bitrate in mind, I will encode to MP3 hoping it will know when to do what with the original audio in order to get a good end-result. The bitrate is not related to quality in my opinion, it is related to the original audio, so the rate should be determined accordingly, and not be tweaked to achieve a certain bitrate average.

I ran a very thorough test on one real-life pop-music track. My goal was to get 3 MP3-files with as close to equal size as possible, using CBR on 192 kbit and a VBR- and ABR-encoded file to match the CBR-size.

Since the VBR file turned out a little smaller than the others, the results were surprising to me. It seems that if you care about the high frequencies, VBR with LAME is the absolute best. Much better even than an even larger CBR-version of the same track!

On the 3 versions I could not really notice very obvious artifacts below 8 kHz, but there was some crap with the S-sounds, noticeable in the ABR version.

I might run some more tests with the newer LAME edition, which has VERY much improved VBR routines working for us. For now:
From what I've heard and seen I don't recommend to use ABR, it's not even competition to the quality you get from VBR.

I just put this page together REAL fast and did the test because I wanted to do a random check to see what the ABR-hype was about. I'm glad I did this(!) and have proof for myself to still prefer LAME's VBR over any other type of MP3 encoding, especially with the recent alpha versions.
I also happen to not really like the concept behind ABR and tried not to be biased in my judgement, but...
Realize I've been testing with just 1 random music track.
On other music it might turn out better for ABR, I simply can't tell yet.
I do know for certain I tested this on a lengthy track (6 minutes) and frequency response peaks over the entire length.

  • See another site with info on Razorlame and how to set it up.

    all L.A.M.E. developers
    Mitiok for the win32 compiles
    Holger Dors for creating RazorLame
    Robert Hegemann for the best VBR
    Our friends @Nullsoft for making
    MP3 what it is today
    Exact CD Copy


    Julius T & Hens Z
    Men Without Plan Enterprises
    FTP server homemade Music and Software