multiband compressor

Hi,

I would like to write a digital audio effect processor with the ADI ADSP-21161N
DSP. I have some troubles to find some references which talk about
compressors and especially about multiband compressors. Can you suggest me
some urls or books which talk about this subject?

I have found during my web surfing that designing a multiband compressor
is very very difficult. Some peoples says that some companies have taken a
lot of years to develop their audio compressors. It seems for me very
exaggerated. What is your opinion?

Thank you.

--
Eric

In article pan.2003.08.03.17.51.25.151778@yahoo.fr, Eric at ralawa@yahoo.fr
wrote on 08/03/2003 13:51:

> I would like to write a digital audio effect processor with the ADI
> ADSP-21161N
> DSP. I have some troubles to find some references which talk about
> compressors and especially about multiband compressors. Can you suggest me
> some urls or books which talk about this subject?
> 
> I have found during my web surfing that designing a multiband compressor
> is very very difficult. Some peoples says that some companies have taken a
> lot of years to develop their audio compressors. It seems for me very
> exaggerated. What is your opinion?

besides choosing the particular bands (there are psychoacoustic reasons for
compressing different frequency bands separately), there is the actually
design of the compressor on each band (assuming that the compressors are the
same structure).  how much delay (or reaction time for the gain adjustment)?
how is the gain adjusted (dB in vs. dB out curve, knee "softness", etc.),
attack and decay times, limiting on the top end, gating on the bottom.  also
make damn sure that the multi-bands line up in delay and frequency response
so that when the compressor levels are set to "no effect" there actually
*is* no effect.

there's probably a slew of issues that i don't know about.

> Thank you.

FWIW,

r b-j

"Eric" <ralawa@yahoo.fr> ha scritto nel messaggio
news:pan.2003.08.03.17.51.25.151778@yahoo.fr...
> Hi,
>
> I would like to write a digital audio effect processor with the ADI
ADSP-21161N
> DSP. I have some troubles to find some references which talk about
> compressors and especially about multiband compressors. Can you suggest me
> some urls or books which talk about this subject?

> I have found during my web surfing that designing a multiband compressor
> is very very difficult. Some peoples says that some companies have taken a
> lot of years to develop their audio compressors. It seems for me very
> exaggerated. What is your opinion?
>
> Thank you.


Hi Eric,
I think the difficulty of making an audio compressor depends on the
psycho-acustic model. This is the most complex part which gives you the
highest compression rates in classic audio compressors like in MPEG audio.
Layer 1 and 2 are sub-band compressors that use 32 subbands and a pseudo QMF
(partial alias cancellation) filter bank. The bit allocation for each
subband is given by a psycho-acustic model. This model work on a finer
spectral decomposition based on FFT.
You could skip the psycho-acustic model with an adaptive bit allocation for
example based on the sub-band energy. This which would reduce the
computational complexity of the encoder and would make it much simpler.
Layer 3 is an hybrid compressor: the subband output is transformed with a
special DCT and this leads to a few other problems like alias reduction in
the time domain.
Other effects concerning maskerading in the time domain are considered from
layer 2.

references
------------
about filterbanks:
P.P. Vaidyanathan, Multirate Systems and filterbanks, Prentice-Hall, 1993

T.Saram&#4294967295;ki and R. Bregovic, Multirate Systems and Filter Banks, Idea Group
Publishing, 2002

T. Karp, A. Mertins, Modulated, Perfect Reconstruction Filterbanks With
Integer Coefficients, IEEE Trans. Signal Processing, vol. 50, no. 6, June
2002



about psycho-acustic models and audio coding:

C. J. B. Moore, An Introduction to the Psychology of Hearing, Academic
Press, 1989

Udo Zolzer, Digital Audio Signal Processing, Wiley, 1997


E. Zwicker, Psychoakustik, Springer-Verlag, 1982



the MPEG 1 audio specifications:

ISO/IEC 13818-3 : Information technology - Generic coding of moving pictures
and audio - Part 3: Audio; Part 7: Advanced Audio Coding



You could send me an e-mail for a deeper discussion on audio coding, I would
be glad.



Emanuele, Italy

Ema wrote:
> 
> "Eric" <ralawa@yahoo.fr> ha scritto nel messaggio
> news:pan.2003.08.03.17.51.25.151778@yahoo.fr...
> > Hi,
> >
> > I would like to write a digital audio effect processor with the ADI
> ADSP-21161N
> > DSP. I have some troubles to find some references which talk about
> > compressors and especially about multiband compressors. Can you suggest me
> > some urls or books which talk about this subject?
> 
> > I have found during my web surfing that designing a multiband compressor
> > is very very difficult. Some peoples says that some companies have taken a
> > lot of years to develop their audio compressors. It seems for me very
> > exaggerated. What is your opinion?
> >
> > Thank you.
> 
> Hi Eric,
> I think the difficulty of making an audio compressor depends on the
> psycho-acustic model. This is the most complex part which gives you the
> highest compression rates in classic audio compressors like in MPEG audio.
> Layer 1 and 2 are sub-band compressors 
<snip>

Compression in this case probably refers to gain reduction, not bitrate/
lossy data compression. There is a class of audio effect processors 
called "multiband compressors" used to tweak audio mixes - "make
it louder". 

SFAIK, multiband compressors divide the singals into bands, 
compress the bands, then mix. Dunno past that. 

--
Les Cargill

Sorry, but I was talking about audio compression like gain control. I was
not talking about compression like mp3 :)

--
Eric

> Compression in this case probably refers to gain reduction, not bitrate/
> lossy data compression. There is a class of audio effect processors
> called "multiband compressors" used to tweak audio mixes - "make
> it louder".
>
> SFAIK, multiband compressors divide the singals into bands,
> compress the bands, then mix. Dunno past that.

Hu hu! sorry about my off-topic ;-)

so are these effect processors so difficult to make? are the the filter
banks different from my case?

E.

Ema wrote:
> 
> > Compression in this case probably refers to gain reduction, not bitrate/
> > lossy data compression. There is a class of audio effect processors
> > called "multiband compressors" used to tweak audio mixes - "make
> > it louder".
> >
> > SFAIK, multiband compressors divide the singals into bands,
> > compress the bands, then mix. Dunno past that.
> 
> Hu hu! sorry about my off-topic ;-)
> 

It's just a common collision of the word "compression". DSP guys
think of them as AGCs, I think, although AGC is a specialization
of the general topic. 

> so are these effect processors so difficult to make?

Dunno. I'm not far enough along the learning curve to say. I'd
think once you had elements of gain reduction and band limiting
figured out, the assembly would be relatively simple. 

My only real DSP exposure is the VST plugin kit. 

>  are the the filter
> banks different from my case?

I don't think they really are. There are fewer bands in these sort of
compressors than say, a .mp3 encoder ( as I understand it ). 

> 
> E.


--
Les Cargill

"Eric" <ralawa@yahoo.fr> ha scritto nel messaggio
news:pan.2003.08.03.20.32.12.215700@yahoo.fr...
> Sorry, but I was talking about audio compression like gain control. I was
> not talking about compression like mp3 :)
>
> --
> Eric

yep! that's clear now :-)

but maybe references about filterbanks could be useful for a fast
(polyphase) implementation of the filtering.

Good luck!
Emanuele

Eric wrote:
> 
> Sorry, but I was talking about audio compression like gain control. I was
> not talking about compression like mp3 :)
> 
Doing multi-band compressors tends to cause some problems with 'phasing'
effects.  You can do it correctly, and it can be made to sound good
(and mitigate some of the problems from single band designs.)  I'll
probably play with multi-band again, but my favorite plaything is still
single-band.  It has no attack/decay time per-se, and is semi-intelligent
about mitigating modulation effects and it never pumps (well, it can
probably be coerced to do so.)  The worst part of my current single
band design is the peak limiter.  Luckily, the compressor is good enough
to minimize the possibility of the limiter 'compressing' the sound, and
it only AGCs on the peaks.   The biggest 'minus' of my single band
design is that it has a significant delay.  However, it is experimental,
and I made the delay long enough to support extremely long attack times.
I suspect that I could easily decrease the delay to 50msec and not cause
significant problems.

There are some single-band tricks that mitigate some of the worst problems,
and simple attack/release times are almost always suboptimal.  If you
calculate the attack/release based upon historical behavior and also the
spectrum/waveshapes, many of the common single band problems can be
mitigated.  (A short transient isn't likely to have alot of low frequency
content, so the attack/decay times can be quicker.  Also, it is a good
idea (if possible) to make gain changes during zero crossings.)

One of my multi-band playthings was based upon an FFT-type approach,
and using a carefully constructed window which mitigates the problems
of overlap.  It sounded very good, but IMO it was too aggressive except
for 'artistic' use.

John

>>>>> "Eric" == Eric  <ralawa@yahoo.fr> writes:

    Eric> I have found during my web surfing that designing a
    Eric> multiband compressor is very very difficult. Some peoples
    Eric> says that some companies have taken a lot of years to
    Eric> develop their audio compressors. It seems for me very
    Eric> exaggerated. What is your opinion?

Making something that compresses is easy.  Making something that
compresses in a way that sounds agreeable to most listeners is harder.
Please do experiment and post any interesting results.

David