MusicMatch 5.10.0149
160kbs, "normal" (joint-stereo)
Sound
Forge 4.5h,
commp3.dll version 1.0, build 219
160kbs, "fastest" (joint-stereo)
Notes by ff123
I identified the FhG codec "FastEnc" in several programs that I looked at: Sound Forge 4.5h (commp3.dll version 1.0, build 219) "fastest," MusicMatch "normal," and Cool Edit Pro 2.1 with the MP3 ME plugin ("fast codec"). In a listening test of the castanets clip, I didn't notice anything unusual at 128kbs using Sound Forge 4.5h as the FastEnc encoder. Imagine my surprise when I listened to the same clip and codec at 160kbs and heard a marked difference from the original. It was as if the reverberation in the clip had been removed!
Addendum (2-9-01): Others have
described the effect as a collapsing of stereo separation (TLO,
David Robinson).
Addendum (2-25-01): Sound Forge 5.0
(commp3.dll version 1.1, build 254) appears to fix the FastEnc
bug.
Addendum (5-09-01): MusicMatch
Jukebox 6.10.0148Beta also appears to have the fixed version (at
long last)
So I went to MusicMatch 5 and listened to a clip generated using that program's implementation of FastEnc. No audible problems. Then I generated new signatures, except now at 160kbs. The plots below show time and frequency plots of both encoders using the last 5 seconds of a standard Sound Forge tone sweep as a source.
| Engine | Time-domain response | Frequency-domain response |
| FhG "FastEnc" codec MusicMatch 5.10.0149 |
|
|
| FhG "FastEnc-bad" codec Sound
Forge 4.5h, |
|
|
Nothing unusual here. The codecs give the same signature, as expected. So why did they sound different? Puzzled, I tried looking at the out-of-phase plots (see my "Out of Phase Quirk in FhG 1.2 build 63i Codec" page for an explanation of these sweeps), as shown below:
| Engine | Time-domain response | Frequency-domain response |
| FhG "FastEnc" codec MusicMatch 5.10.0149 |
 |
 |
| FhG "FastEnc-bad" codec MusicMatch
6.00.0270 |
 |
 |
| FhG "FastEnc-bad" codec Sound
Forge 4.5h, |
 |
 |
So FastEnc as implemented by Sound Forge 4.5h is doing something unusual which I can see as well as hear (sigh of relief). Also, the signatures for FastEnc in Sound Forge 4.5h and MusicMatch Jukebox 6.0 look similar (see my signatures page), so I expect that MusicMatch Jukebox 6.0 has the same problem (note: MusicMatch 6.1 Beta has fixed the problem).
Almost all of the Sound Forge 4.5h out-of-phase signal seem to be attenuated about 8dB compared with MusicMatch 5. But why didn't I hear it at 128kbs? I don't know. Whatever the case may be, the bottom line is that Sound Forge 4.5h has the same shape at 192kbs and 256kbs that it does at 160kbs, and considering the fact that at 160kbs I heard a clear difference that shouldn't be there, I don't recommend using either Sound Forge 4.5h ("fastest") or MusicMatch Jukebox 6.0 ("normal"). Also, since VBR in these programs is based on FastEnc, I don't recommend using VBR in these programs either.
| Engine | Time-domain response | Frequency-domain response |
| FhG "FastEnc" codec MusicMatch 5.10.0149 |
 |
 |
| FhG "FastEnc-bad" codec MusicMatch
6.00.0270 |
 |
 |
| FhG "FastEnc-bad" codec Sound
Forge 4.5h, |
 |
 |