MPEG Layer III: which bitrate for 32 kHz audio?

MPEG Layer III: which bitrate for 32 kHz audio?

Post by Norbert Weinri » Wed, 15 Oct 1997 04:00:00



Could somebody please advise me which bitrate I should use for
encoding 32 kHz audio?

I understand that for 44.1 kHz a bitrate of 112 kbit is supposed to
yield "CD-quality". Does this mean that (due to the lower sampling
rate and to the therefore fewer audio information) the optimal bitrate
for 32 kHz audio would be 96 kbit or do I have to choose 112 kbit in
both cases?

Thanks for your advice,
Norbert

 
 
 

MPEG Layer III: which bitrate for 32 kHz audio?

Post by ET3 Ba » Thu, 16 Oct 1997 04:00:00



Quote:(Norbert Weinrich) writes:
>Could somebody please advise me which bitrate I should use for
>encoding 32 kHz audio?

>I understand that for 44.1 kHz a bitrate of 112 kbit is supposed to
>yield "CD-quality". Does this mean that (due to the lower sampling
>rate and to the therefore fewer audio information) the optimal bitrate
>for 32 kHz audio would be 96 kbit or do I have to choose 112 kbit in
>both cases?

>Thanks for your advice,
>Norbert

As with any sample-rate, it depends on the content. If it is Orchestral
music which has a lot of content to mask compression artifacts, you could
get away with 96kbit/s, but if it is more minimalist (like vocals with
guitar and bass) even 128kbit/s will sound a little 'squishy'.

The lower samplerate only reduces the upper frequency limit from ~20kHz to
~16kHz, this high frequency information is usually removed during
compression since normally, it is relatively inaudable (perceptual
compression) - now it simply isn't there. So you still need about the same
datarate.
One advantage though, 32kHz audio decoding requires <1/2 the processing
time for a given datarate as compared with 44.1kHz.

Stede Bonnett   ***noted ten BeSt***  -br 128000 -crc -hq
http://members.aol.com/et3bahn/mp3/mp3.html

 
 
 

MPEG Layer III: which bitrate for 32 kHz audio?

Post by Bjor » Thu, 16 Oct 1997 04:00:00



>>Could somebody please advise me which bitrate I should use for
>>encoding 32 kHz audio?

>As with any sample-rate, it depends on the content. If it is Orchestral
>music which has a lot of content to mask compression artifacts, you could
>get away with 96kbit/s, but if it is more minimalist (like vocals with
>guitar and bass) even 128kbit/s will sound a little 'squishy'.

I would give the completely opposite advice. MPEG compression builds on
psychoacoustic theory just like noise suppression, but it definately doesn't
have the same characteristics. If there is something that MPEG-audio has
trouble compressing into a given bitrate, it's orchestral stuff and
everything else that has a broad loud spectral contents like choirs - most
of the spectra is so important that you can't really add noise (compress)
anywhere (exaggerating a bit here).  You could put in unaudible noise in the
few silent energies like you suggest, but it really isn't enough.

On the other hand, a bass and a vocal is compressed very very nicely using
MPEG compression, because the spectra is so scarce and free of transients
that the available bits can be used to quantize the existing energies less
coarsely.

The mpeg compression artefacts are easily distinguishable in all
compressions of orchestral music I've heard, but they are much harder to
hear in traditional rock.

Quote:>The lower samplerate only reduces the upper frequency limit from ~20kHz to
>~16kHz, this high frequency information is usually removed during
>compression since normally, it is relatively inaudable (perceptual
>compression) - now it simply isn't there. So you still need about the same
>datarate.

It is true that if there is no spectral information in the upper frequency
regions, then you would not really get any gain from using say 48khz instead
of 32khz. On the other hand, if there IS spectral audible content, then it
would sound bad if you cut and use 32khz samplerate. If you have no other
option though that is the way it has to be.

So why don't you just compress whatever you have in both 96 and 128 kbit and
listen to see what sounds best? After all, there is no way to objectively
specify the performance of psychoacoustical compression schemes. All
"specifications" arise from subjective listening-tests.

/Bjorn

 
 
 

MPEG Layer III: which bitrate for 32 kHz audio?

Post by jj, curmudgeon and tiring philalethi » Thu, 16 Oct 1997 04:00:00



>On the other hand, a bass and a vocal is compressed very very nicely using
>MPEG compression, because the spectra is so scarce and free of transients
>that the available bits can be used to quantize the existing energies less
>coarsely.

Sometimes true, sometimes wrong.

Quote:>The mpeg compression artefacts are easily distinguishable in all
>compressions of orchestral music I've heard, but they are much harder to
>hear in traditional rock.

Um, "all compressions of orchestral music"? Which MPEG layer are you
using. It's been my experience that most orchestral music sounds pretty good
at 128kb/s for layer 3, and better for AAC, and that it's NOT a critical
source. The choices of critical material for MPEG testing also suggest
that this is the case.

Many materials of all sorts are searched for artifacts during the
testing, and orchestral stuff does NOT make the final cut. I've been
there most of the time since about 1989, and I'm quite sure
of this.

Quote:>After all, there is no way to objectively
>specify the performance of psychoacoustical compression schemes.

I'll go along with that, 10/4 to the contrary.

--
Copyright alice!jj 1997, all rights reserved, except transmission by USENET
and like facilities granted.   This notice must be included.  Any use by a
provider charging in any way for the IP represented in and by this article
and any inclusion in print or other media are specifically prohibited.

 
 
 

1. MPEG layer II, III and Real Audio 3.0: what is best?

There are some points which are not clear for me:

How many persons took part in this listening test? Is this only one?
how many test items (= test sounds) are tested here ? Is it only
'Cembalo'?

I think if one want to make any predications on sound quality of
percepual coders, these predications can only be statistical ones,
since every listener would hear other artefacts on different test items.

But for a statistical analysis, you need more than one listener and one
test item, isn't it ?
(For a good description on how to perform and evaluate listening test,
see: ITU Recommendation ITU-R BS.1116)

There seems to be some  confusion  about the terms 'transparent'
and 'non-transparent'

The 'transparentcy' does not necessarily  depend only on the audio bandwidth.

There is no *absolut* 'transparent' or 'not transparent' codec system.
Transparent means that *one* listener is not able to distinguish between
the original of a *certain* test item  and the 'falsification'
(= encoded and decodes signal).
In other words: You  can merely say that one  codec is  transparent for
X percent of listeners for Y percent of test items.

And moreover test have shown:

Even if one test person is able to hear a single 16,5 kHz tone, most
of those testpersons are not able to hear any difference for most
'normal' test items ( pop, klassic,speech) if they shall compare the
original pcm file  with a 16kHz band-limited pcm file.

Means: A lowpass at 16kHz is *not* transparent for a sine tone at 16,5
kHz but *is* transparent for most other testsignals.

Nevertheless,  the predication that Layer3 is better then RealAudio 3.0,
which is better than  Layer2 points to the right direction,
as far as I know.

Bodo Teichmann
--

  Fraunhofer Institut fuer Integrierte Schaltungen (FhG/IIS)
Tel.: +49 9131 776367                  Fax.: +49 9131 776399 */

2. From field of msgs from su'ed id.

3. MPEG Audio Layer III player for Windows

4. Request for Mac D/A converter

5. MPEG audio layer III

6. ZoneAlarm problems...

7. MPEG Audio Compression (Layer III)

8. MPEG 1 layer II vs layer III

9. Mpeg layer III

10. MPEG Layer III

11. MPEG 1 Layer III