Time and Frequency Domain Signatures of
Various Codecs
Notes by ff123
The various encoders can be identified by looking at the
output they produce in the time and frequency domains. I
generated a stereo, 10 second test tone-sweep (4.5kHz to 22kHz).
The last 5 seconds of this sweep were encoded with various
engines, then decoded back to .wav format and converted to mono
to increase the vertical resolution of the time-domain graphs (L
and R channels were identical in all cases, except for the FhG
codec found inside Liquifier Pro 5.0). The file halfsweep.flac is
available from my
Audio Samples Page. It has been compressed using the lossless
compressor, FLAC. Tone
sweeps generated with Sound Forge cause problems for codecs (see
my MP3ENC clicking noises page). This
makes them ideal for creating signature plots. It turns out
though, that signatures made from sweeps aren't quite enough to
distinguish one codec variation from another. So to further
discriminate variations among the broader codec families, I
encoded a short clip of applause at 128kbs (you can download this
clip from the Lame
Quality and Listening Test Information site).
Please note that I advocate subjective listening tests of real
music to evaluate codec quality. One might be tempted to
hypothesize that one codec should sound better than another
because it has a better-looking response, but listening tests are
the real proof of the pudding!
Thanks to debtman7 and to TLO (web page here)
for providing me with the MacOS mp3 signature files.
There are a variety of FhG mp3 engines:
1. Version 1.0 build 63i, found in programs such as Opticom .mp3
Producer Pro 2.1 as a program-internal codec, and in a hacked
distribution ("Radium") as an external (ACM) codec,
where it has a version of 1.2 build 63i. There is a hacked
version of .mp3 Producer Pro, which is billed as "freeware"
and is sometimes referred to by users as L3 Producer Pro.
Producer Pro's internal codec encodes only with joint-stereo at
192kbs. The hacked external codec allows the option of choosing
the stereo mode and encoding with bitrates of 224 and 320. Plus
it's faster.
2a. "Alternate-bad" codec, found in such applications
as Cool Edit and MusicMatch Jukebox 5
2b. "Alternate" codec, found in Sound Forge 4.5h (commp3.dll
version 1.0, build 219) and MusicMatch Jukebox 6 -- same as
"Alternate-bad" except it doesn't crash or produce bad-looking
output with tone sweeps; it sounds different as well)
3a. "FastEnc" codec, found in MusicMatch 5 and Cool
Edit Pro with the MP3 ME plugin (but not in Cool Edit 2000)
3b. "FastEnc-bad" codec, found in Sound Forge 4.5h (commp3.dll
version 1.0, build 219) and MusicMatch Jukebox 6 -- same as
"FastEnc" except that it has a bug in it which
attenuates the out-of-phase signals, resulting in a very "dry"
sound.
4. MP3ENC 3.1 codec (September 23, 1998)
5. MP3ENC 3.1.1 codec (April 28, 2000, not distributed to the
general public; that includes me). Note; Ivan Dimkovik, developer
of Psytel AAC, writes: "MP3enc 3.1.1 is
actually a name for FhG MP3 library, version 3.1.1 (dated march,
2000). This library has both CBR and VBR modes and it is built
into many software packages. It is possible to get the library by
signing NDA with Fraunhofer." So it is my guess that
this library contains the codecs referred to above as "Alternate
and FastEnc."
6. Liquid Audio codec
Click on thumbnail images to zoom.
Alternate Codec: The half-sweep plots show that these all
belong to the same codec family. Fraunhofer has evidently updated
and improved the sound of this codec, where it is now used in
programs such as Sound Forge 4.5h and MusicMatch Jukebox 6.
This is the middle setting in each of the
programs listed. The half-sweep plots are similar enough to the
ones above to make me believe that this is in the same "Alternate"
codec family. In Cool Edit, this setting is termed "fast,"
but it is really just quality setting 3, further supporting this
idea.
MP3Enc and the Alternate codecs are related to
each other, I think. Both codecs are tripped up by the same
difficult-to-encode pieces of music, and both take a very long
time to encode at their highest quality settings. MP3Enc and
Alternate are the only codecs I've heard to produce "dropouts"
as artifacts (see the Lame test
samples site under the BlackBird.wav entry).
| Engine |
Time-domain response,
half-sweep
192kbs (stereo) |
Frequency-domain response
half-sweep
192kbs (stereo) |
Frequency-domain response
Applaud clip
128kbs (joint stereo) |
MP3ENC 3.1,
qual9 |
 |
 |
 |
The lowest setting in Sound Forge ("fastest")
is the same as the lowest setting in MusicMatch ("normal").
Cool Edit 2000 doesn't seem to have a quality setting which
produces a signature that looks like this, but Cool Edit Pro with
the MP3 ME plugin does. On the applaud clip, the Sound Forge 4.5h
and MusicMatch Jukebox 6 implementations of FastEnc looks
markedly different from all of the other codecs (See my page, Bug in latest FhG FastEnc).
The FhG mp3 codec within Liquifier Pro does not
produce repeatable encodes with halfsweep.wav, even within the
same session. This is apparently the result of watermarking (?)
Shown below are two encodes of halfsweep.wav using the exact same
settings.
| Engine |
Time-domain response,
half-sweep
192kbs (stereo) |
Frequency-domain response
half-sweep
192kbs (stereo) |
Frequency-domain response
Applaud clip
128kbs (joint stereo) |
LAME 3.87alpha with
RH extensions,
hq |
 |
 |
 |
| Engine |
Time-domain response,
half-sweep
192kbs (stereo) |
Frequency-domain response
half-sweep
192kbs (stereo) |
Frequency-domain response
Applaud clip
128kbs (joint stereo) |
Xing plugin for
AudioCatalyst 2.1 |
 |
 |
 |
| Engine |
Time-domain response,
half-sweep
192kbs (stereo) |
Frequency-domain response
half-sweep
192kbs (stereo) |
Frequency-domain response
Applaud clip
128kbs (joint stereo) |
| Qdesign MVP 1.2.51 |
 |
 |
 |
| Engine |
Time-domain response,
half-sweep
192kbs (stereo) |
Frequency-domain response
half-sweep
192kbs (stereo) |
Frequency-domain response
Applaud clip
128kbs (joint stereo) |
| SoundJam MP 2.5.2 for Mac OS (Best quality, enabled
Smart Encoding Adjustments, filter frequencies below 10
Hz) |
 |
 |
 |
| iTunes 1.0 for MacOS (high quality default setting |
 |
 |
 |
Return to ff123's home
page
