Bug in older FhG "Alternate" Codec; Also reports of bad sound

Notes by ff123

Cool Edit 2000 in the quality 9 setting (FhG "Alternate" codec) will crash when encoding a standard 10 second tone sweep generated by Sound Forge (-0.4dB FS). I tried three bitrates: 256kbs, 192kbs, and 128kbs. David Robinson reports similar problems using Cool Edit Pro 2.1 with the MP3 ME plugin.

To try to keep the programs from crashing, I generated a tone sweep at -12dB FS and deleted the first 5 seconds. This time Cool Edit 2000 didn't crash, but the output produced was clearly wrong, showing hard clipping beginning at 14.6kHz. For the same 5-second half-sweep, MusicMatch 5.10.0149 produced similar output in its "very high" setting (to be expected since it uses the same FhG "Alternate" codec; see my signatures page). This bug has been noted by Info2 on his r3mix site (see analysis section), who says that the same behavior is also apparent in Nero 5 and Proteron EasyMP3 v1.00.

Sound Forge 4.5h and MusicMatch Jukebox 6 do not have a problem encoding this tone sweep at any of the three bitrates I tried. Fraunhofer has evidently updated this codec.

This problem is dependent on the input fed to the codec. Tone sweeps generated by Sound Forge tend to give codecs problems because the frequencies are synthesized rather than ramped (see my MP3ENC clicking noises page). No problems are encountered if a ramped sweep is used.

Click on the thumbnail images to zoom.

FhG Engine	Time-domain response	Frequency-domain response
Cool Edit 2000 v1.0, 256kbs, qual9
Music Match 5.10.0149, 256kbs, "very high"
Sound Forge v4.5h, 256kbs, "highest quality"

What about the sound? Sure, software shouldn't crash, but does software that crashes make for bad-sounding encodings? Here's what others have to say about how the MusicMatch 5 "very high" or "high" (Alternate) codec sounds:

2Bdecided (original post on the archived MP3.com BulletinBoard):
Trust me - the LAST thing you want to use for music is the "Alternative codec (High)"/"Very High" (different applications, different names!). I've never heard anything worse at 256kbps. (apart from maybe the GNU mp2 CEP plug-in at a slightly lower bitrate - now that was horrible!)

bAdDuDeX (original post on the VQF.com Users' Forum):
Any song encoded on High or Very High sounds horrible to me. It seems like people who can't hear high frequencies well don't notice all the artifacts that occur when using that codec (Normal quality uses a different codec if you didn't know). I still think MP3Enc is the best MP3 encoder.

bAdDuDeX (original post on the VQF.com Users' Forum):
The artifacts sound like "ringing" or something like that. It's hard to describe. They're the same type of artifacts that occur with BladeEnc and LAME (less with LAME though). I still haven't figured out what causes it... It's probably either frequency vibrations or that the encoders allocate a very small amount of bits for high frequencies. But with that codec, songs sound horrible to me even at 192kbit/s with artifacts throughout. It's one of the worst codecs I've ever heard, I'd even take Xing over it.

bAdDuDeX (original post on the VQF.com Users' Forum):
I still think it sounds horrible at 128kbit/s. It gets worse in comparison with other encoders as the bitrate goes up. At 192kbit/s, I might even say it IS the worst encoder. I never even tried 256kbit/s because it sounded too horrible at 192 already.

Adam (original post on the VQF.com Users' Forum):
It sounds like a watery ringing. When you encode in "Very High" it sounds like the bandwidth was unnaturally increased and that it all went to "flange".

All this prompted me to design a short test to compare Lame 3.87a (RH) against MMJB 5 "very high" at 192kbs. This is the bitrate where Lame should be coming into its own, but also where the nasty effects of the FhG codec should be very audible. I chose to encode Applaud.wav (which can be found at the Lame Test Samples site) for people to listen to. The files can be found in my Audio Samples Page. They have been compressed using the lossless compressor, FLAC. There are three files, as shown below. Can you tell the difference between either of the two encoded files and the original? Can you tell which one sounds worse (and is presumably the one encoded by MusicMatch 5)? My own results are documented in my listening tests of applaud. Although I could identify both files from the original, I could not say that one codec sounded better or worse than the other.

bAdDuDeX (original post on the VQF.com Users' Forum):
XX is MMJB. I can't mistake that sound. [note, I have obscured bAdDuDeX's correct result, in case you'd like to try this test for yourself].

HansHeijden (original post on the MP3.com Message Board):
ApplaudXX is terrible, it has an extra sound as if someone is also waving with a sheet of aluminium foil. This sound is well below 16 kHz I think. I have little doubt about the encoder used...

The two others sound the same to me.

 

Next, I asked people to listen to a file encoded with Sound Forge's implementation of the "Alternate" codec. This is in the same codec family (see my signatures page), but it is different. Since MusicMatch 6 yields a similar signature, I expect that it will sound the same.

bAdDuDeX (original post on the VQF.com Users' Forum):
Wow, the Sound Forge one sounds MUCH better. It doesn't have all those annoying artifacts that are normally there. I looked at the waveform in Cool Edit and it has a 16kHz cutoff with spikes to 20kHz, which is different from the normal Alternate codec that has a 14kHz cutoff with spikes to 20kHz. So it seems like the spikes are the thing that actually cause the artifacts! That's what one of my theories was but it sounded pretty crazy... I have little doubt now that if your ears can't hear high frequencies well, then you won't hear the artifacts.

I plotted spectral views of various codecs using applaud.wav at 192 kbit/s to show what is happening in frequency over time. The older Alternate codec (MusicMatch Jukebox 5) has a primary low-pass cutoff of 13 kHz instead of the normal 16 kHz. Higher frequencies appear often, but intermittently, most likely causing the "waving sheet of aluminum foil" sound which Hans Heijden describes. The newer Alternate codec (Sound Forge 4.5h) cuts off at 16 kHz and higher frequencies appear much less often. FastEnc produces high frequency excursions still less often, and MP3Enc and FhG 1.263 never exceed 16 kHz. This is remarkable at 192 kbit/s -- encoders are easily capable of producing output above 16 kHz at this bitrate. The mp3 codec within Liquifier Pro 5.0 is unusual and most likely sounds bad -- it has itermittent but highly amplified (about 12 dB) output between 16 and 18 kHz.

Encoder	Spectral view of applaud.wav
Original
Alternate codec MusicMatch Jukebox 5.10.0149, very high
Alternate-Aug 2000 codec Sound Forge 4.5h, highest quality
Lame 3.87RH -h
FastEnc (Cool Edit Pro with MP3 ME plugin, fastest, CRC disabled)
MP3Enc31, -qual 9
Fhg 1.263i, joint stereo
Liquifier Pro 5.0 mp3

Return to ff123's home page