Notes by ff123
Listening test conditions: Grado SR325 headphones, fanless computer operation, mp3-to-wav conversion performed with the MAD 0.12.3 beta 4 plugin for Winamp.
I performed a listening test using fatboy.wav (found on the Lame test samples page) to try to rank FhG VBR with the CBR implementation. Fatboy.wav is ideal for this test because artifacts are audible even at very high bitrates.
I used the FhG FastEnc implementation found in Cool Edit Pro with the MP3 ME plugin. The newer version of FastEnc, as found in MusicMatch Jukebox 6, is buggy (see my page, "Bug in Latest FhG FastEnc"), and in both the previous and current versions of MusicMatch, the VBR implementation cuts off at 16 kHz, even with 100% VBR quality. [Addendum, 2-5-01]: Max bandwidth > 16 kHz can work in MusicMatch 6, but only if you specify the max bandwidth manually, and only if you create an mp3 directly from CD!
Cool Edit Pro with the MP3 ME plugin has twoVBR presets, "High Quality Stereo" and "Medium Quality Stereo." The specific default settings chosen for the High Quality Stereo preset are shown in the screenshot below:
The Medium Quality Stereo preset will choose 50% VBR quality instead of 100%, and the Maximum Bandwidth will be set to 16500 Hz instead of to 20000 Hz.
It is also possible to use the VBR quality pulldown menu to choose VBR settings. For example, if I choose the VBR quality setting of 70%, I get the following (see screenshot below):
By selecting 70 from the pulldown menu, I have overridden the High Quality Stereo preset. Maximum bandwidth is set to 19170 Hz, and the VBR quality, is of course set to 70%. If I pull down 100% quality, I will get a different maximum bandwidth than if I use the High Quality Stereo preset. For the 100% and 50% VBR quality settings, therefore, I chose the presets rather than the pulldowns. I didn't notice any difference in sound by doing so.
The table below shows the settings I used for both CBR and VBR. Note that I disabled the CRC checksums in each case. CRC checksums take away slightly from the amount of bits allocated to the encoder.
| CBR or VBR | Bitrate | Quality | Stereo mode | Maximum Bandwidth (Hz) | CRC Checksum |
| CBR | 128 | N/A | joint stereo | 15500 | no |
| CBR | 160 | N/A | joint stereo | 20129 | no |
| CBR | 192 | N/A | joint stereo | 21000 | no |
| CBR | 256 | N/A | joint stereo | 22050 | no |
| VBR | 134 | medium (50%) | joint stereo | 16500 | no |
| VBR | 142 | 60% | joint stereo | 18380 | no |
| VBR | 169 | 70% | joint stereo | 19170 | no |
| VBR | 192 | 80% | joint stereo | 20050 | no |
| VBR | 218 | 90% | joint stereo | 21000 | no |
| VBR | 253 | high (100%) | joint stereo | 20000 | no |
The following rankings are subjective, that is, while I'm able to tell that they each sound different from one another (except as noted), the judgment of which one sounds better than the other is a matter of opinion. Also, don't forget that I can't hear high frequencies well, so it's possible that 70% VBR is getting most of its benefit at the expense of high frequencies, and I wouldn't be able notice the loss.
| Rank | Setting | Bitrate | Comments |
| 1a | 100% VBR | 253 | too close to 256kbs CBR to call |
| 1b | 256kbs CBR | 256 | see above |
| 2 | 90% VBR | 218 | just slightly worse than the above two |
| 3 | 80% VBR | 192 | |
| 4 | 70% VBR | 169 | definitely better than 192kbs CBR |
| 5 | 192kbs CBR | 192 | definitely better than 60% VBR |
| 6 | 60% VBR | 142 | slightly better than 160kbs CBR |
| 7 | 160kbs CBR | 160 | slightly better than 50% VBR |
| 8 | 50% VBR | 134 | definitely better than 128kbs CBR |
| 9 | 128kbs CBR | 128 |
From the ranking above (subject to my caveats), I'd say the sweet spot is probably 70% VBR, which produced a significantly smaller file than 192kbs CBR, while sounding significantly better.
Addendum Mar 1, 2001: EarGuy's digital ear (see my page, Analytical Model of the Ear) agrees with my comparative ranking!