Mp3 Codecs Not Recommended

Notes by ff123

This is a list of mp3 codecs which, for whatever reason, I would not personally use. You can identify a codec by comparing its signature against the list I've compiled on my signatures page. The table at the bottom shows how these (non) recommendations apply for some popular programs.

1. Fraunhofer older "Alternate" codec, as found in MusicMatch Jukebox 5 ("high" or "very high" settings). This codec actually sounds good to me, but other people hear horrible high-frequency artifacts. See my page "Bug in older FhG Alternate Codec." Other programs which use this flawed codec are Nero 5.10, Proteron EasyMp3 1.1.0, Cool Edit (2000, Pro, and Pro with the MP3 ME plugin). The recommendation against using this codec goes against the generally good advice of evaluating quality with one's own ears. But unless you can hear high frequencies well, you may not notice what's wrong with it.

Fraunhofer has updated this codec -- the newer version is available in MusicMatch Jukebox 6 and in Sound Forge 4.5h. This version is much better than the older version, according to one person who could hear the nasties in the older version. The "high" setting in MMJB 6 is a good compromise between speed and quality.

NOTE: The fast codec found in the programs listed above is not the Alternate codec. It is FastEnc. Fraunhofer packaged two different codecs into the same package which it licensed to others. Exception: Cool Edit 2000 does not include FastEnc.

2. Fraunhofer newer "FastEnc" codec, as found in MusicMatch Jukebox 6.0 ("normal" setting). Fraunhofer introduced a bug which tends to make encoded files sound dry, or lacking in reverberation (see my Bug in Lastest FhG FastEnc page). Another program which uses this flawed codec is Sound Forge 4.5h.

The older version of this codec, available in MusicMatch Jukebox 5 and in Cool Edit Pro with the MP3 ME plugin, sounds fine.

Addendum (02-25-01): Sound Forge 5.0 with commp3.dll version 1.1 (build 254) has fixed the bug in FastEnc.
Addendum (05-09-01): MusicMatch Jukebox 6.10.0148Beta has the fixed version as well.

3. Fraunhofer VBR based on newer "FastEnc" codec, as found in Music Match Jukebox 6.0 (version 6.1 Beta is ok). Fraunhofer's VBR is based on FastEnc. Thus, if FastEnc is broken, VBR will be broken as well. Another program which uses this flawed codec is Sound Forge 4.5h. The older VBR is fine, except as noted below.

4. Fraunhofer VBR (based on older or fixed FastEnc) at quality levels approaching 100%, specifically as found in MusicMatch Jukebox 5. VBR in MMJB 5 is fine, as long as you don't set your expectations too high concerning the 100% VBR quality setting. MMJB VBR limits the bandwidth to 16 kHz, even at 100%. What one should see (but doesn't) is an increase bandwidth as VBR quality is increased towards 100% (see my page FhG VBR). So for the extra file size, you're not getting the very high frequencies in return. Some people may not mind, but others will.

In contrast, the VBR implementation found in Cool Edit Pro with the MP3 ME plugin does increase bandwidth as VBR quality approaches 100%. But if you use MMJB 5 instead, I would stick with 50% quality VBR.

MMJB 6.0 does allow for bandwidth control, but only when specifying the maximum bandwidth manually, and only when recording directly from CD! But since FastEnc is broken in MMJB 6.0, this is moot.

5. Xing at 128 kbit/s, as found in AudioCatalyst 2.1. The sound quality of files produced with Xing at 128 kbit/s leaves much to be desired when compared against the better mp3 codecs. The higher the bitrate, the less pronounced is the difference in quality.

6. QDesign MVP 2.1 at any bitrate. On the files I tried, QDesign ranked at or near the bottom. In one clip (castanets.wav), the sound got worse at very high bitrates. (see my listening test of castanets.wav).

7. Blade at 128 kbit/s. At 128 kbit/s, Blade is even worse than Xing. Gabriel Bouvigne lists the encoder limitations which contribute to the almost unbelievably bad quality of Blade at 128 kbit/s:

No lowpassing
wrong transient detection
too late short block switching
bad reservoir handling
no joint stereo
non optimal huffman compression
a lot of bugs everywhere
wrong ath
bad best quantization selection
no scalefactor use
no scfsi

Return to ff123's home page