[M4IF Technotes] Suggest some MPEG 4 Encoders

[Bernhard Grill]

Stephen McGrath wrote:
> Hello, everyone.
> 
> I beg to differ on the statements made here about lossy codec quality,
> particularly about AAC at 128k. Subject to the usual provisos about what
> you are looking for, quality of playback system, etc. etc., there is
> definitely a noticeable difference between source and AAC at 128 or even
> 256.
> 
> Please remember we are talking about audiophile quality here - this was
> the context of the original question. For mass market consumption,
> listening over a mass market PC soundcard with cheap speakers, etc. -
> sure, the output is indistinguishable. But if you invest in a good audio
> playback system, then there is absolutely a noticeable difference. I am
> not claiming that everyone will hear it, but I am claiming that trained
> listeners and/or audiophiles will. "Trained listener" here does not
> necessarily mean someone who does this professionally, just someone who
> has listened often enough that they come to realize what they are
> hearing and recognize the distortions.
> 
> Which is fine, there is nothing wrong with this. This is why there is a
> much larger market for $400 bookshelf audio systems than for audiophile
> gear. To each his needs. Just don't claim there is no difference!
> 
> I can personally back this from 15 years of working in speech and audio
> compression, and much longer being an audiophile (which appears to be an
> incurable disease!) And yes, I have done this through double-blind
> testing. Many times. Repeatably. :-)

Hallo,
at 256 kbit/s I'd be interested on what experience the word "definitely" is 
based on. Even at 128 kbit/s it is extremely difficult to find a few critical 
items where trained listeners in a perfect environment can reliably detect a 
difference (see MPEG Document N2006 "Report on the MPEG-2 AAC Stereo 
Verification Tests";). This of course assumes that you are using a very good 
encoder, like e.g. the Fraunhofer encoder used in this test :-).
A more general remark:
All the audio sources out there today have some limitations. Most of these are 
sometimes more severe than the effects of AAC coding at a reasonably high 
bitrate, e.g.:
CD-Audio:
An Audio-CD contains coded music and is lossy. The resolution is limited to
16 bit accuracy at 44.1 kHz. AAC on the other hand can do 48, 96, or 192 kHz 
sampling rate and is capable to handle dynamic ranges better than 24 bit. So if 
I'd look for suitable material I'd definitely find that AAC or also MP3 (48 kHz 
sampling rate, better than 24 bit dynamic range) do much better than CD-Audio 
for such items.
Analog Vinyl Records:
The dynamic range is not really impressive and then just consider that low 
frequencies are attenuated and high frequencies amplified on the records. The 
inverse filters in the players/amplifiers only roughly invert this process. None 
of the filters involved in this process have linear phase, so compared to the 
original there are phase distortions, etc.
Analog Tapes:
I remember something I did in the early days of MP3. Somebody asked me to send a 
demo. At this time PCs weren't able to decompress mp3 in realtime and CD-Roms 
were also almost unknown. The person also didn't have a DAT, so the only 
remaining possibility was to send an analog tape (OK, it only was a compact 
casette tape). I used a very expensive studio machine. Listening carefully to 
the recorded music I noticed there were audible drop-outs on the relatively 
short sequence. I made several recordings, changed the tape, including trying a 
second vendor and also tried a second recorder, but I never got a recording 
without an audible dropout in our relatively perfect listening environment
Another thing:
Also amplifiers, loudspeakers, (cables -;) ), etc alter the sound (especially 
the very expensive tube amplifiers which often add an audible hum and other 
noises to the original sound, not speaking of the slight non-linearities).
Altogether AAC has shown a remarkably constant performance over a wide range of 
different input signals in several independent tests, if good encoders were 
used. And this is the real tricky part in audio coding: To achieve good 
performance over the whole range of input signals. All the MPEG-Audio Standards 
have shown over time that they were designed carefully in this respect.
Best regards,
Bernhard Grill
-- 
Dr. Bernhard Grill                                  email: grl   iis.fhg.de
Head of Audio Department, FhG-IIS A                 phone: +49 9131 776-351
Am Weichselgarten 3, D-91058 Erlangen, Germany      FAX:   +49 9131 776-398