Mp3 Algorithm

Started by Jacobs, Ben February 16, 2001
Hi

Where can I find Mp3 compression algorithms ?

Ben
	
Hi!

"Jacobs, Ben" wrote:

>                 > Where can I find Mp3
compression algorithms ?
> 
>                 What do you are looking for: information or source code?
> 
>                 Alexander
> 
>                 I want to understand the compression technique in detail,
> but also apply it. Thus both info and source code will be welcome.
> 
>                 Ben

For understanding, you can find some papers and information at
the official MPEG homepage
http://www.tnt.uni-hannover.de/project/mpeg/audio/
and
http://www.mp3-tech.org

For source code, the lame project should be most interesting,
I'm not sure if the links are correct:
http://www.mp3dev.org
http://www.sulaco.org/mp3

Hope this helps somewhat,
Alexander

-- 
alexander lerch
lerch@lerc...

zplane.development
holsteinische str. 39-42
12161 berlin
 fon: +49.30.854 09 15.0
 fax: +49.30.854 09 15.5
	
Hi,
  You can go to www.mpeg.org and its gives many links to MP3 papers,
software etc. You can buy standard from ISO site.
Regards
Mihir Mody
	On Fri, 16 Feb 2001, Jacobs, Ben wrote:

> Hi
>
> Where can I find Mp3 compression algorithms ?
>
> Ben
>
>
>
> _____________________________________
> Note: If you do a simple "reply" with your email client, only the
author of this message will receive your answer.  You need to do a "reply
all" if you want your answer to be distributed to the entire group.
>
> _____________________________________
> About this discussion group:
>
> To Join:  audiodsp-subscribe@audi...
>
> To Post:  audiodsp@audi...
>
> To Leave: audiodsp-unsubscribe@audi...
>
> Archives: http://www.egroups.com/group/audiodsp
>
> Other DSP-Related Groups: http://www.dsprelated.com
>
>
	
HI Jeff,
   When you buy CD from standard, in CD they give some conformance
streams. Your decoder should able to decode and final output error
should not exceed the error specified from reference file. The exact
procedure for testing of decoder using mp3 file is specified in part4
of standard (conformance testing).
  For encoder only restriction is that the bitstream should confirm
the standard. The quality of encoded file depends on Encoder
implementation ( quantizor, psycho-acoustics etc). So output encoded
files of MP3 encoder are of different audio quality depending on
the encoder implemetation, But decoder will able to decode all
mp3 files. For high quality encoder, you have to pay more.
Regards
Mihir Mody
Saken Communication technology Ltd
	On Mon, 19 Feb 2001, Jeff Brower wrote:

> Mihir Narendra Mody-
>
> >  You can go to www.mpeg.org and its gives many links to MP3 papers,
> >software etc. You can buy standard from ISO site.
>
> One problem we have with calling the ISO site "standard":  there
are no standard
> input and output vectors to use for bit-exactness verification.  I.e. there
are
> no given .wav (or .aif) input files, .mp3 output files, and resulting
decoded
> .wav files for comparison.
>
> We also tried to get comparison files from Fraunhofer, even going so far as
to
> contact upper management.  Nothing.  Fraunhofer has some .mp3 files on
their
> site, but they appear to withhold the original .wav files.
>
> So everyone's implementation of MP3 is possibly -- likely at least somewhat
--
> different, as long as it sounds Ok?  I guess so.
>
> Any comments on this issue are greatly appreciated.
>
> Jeff Brower
> DSP sw/hw engineer
> Signalogic
>
	
Hello Jeff Brower,

Jeff Brower wrote:
> 
> Mihir Narendra Mody-
> 
> Thanks for your informative reply.  It's more than I thought was there, but
> still too imprecise.  There should also be conformance streams for the
encoder,
> given certain parameters.

In this case, it wouldn't be possible to improve such codecs
in quality. This would be not very good, since all first
generation audio codecs only provided a quite bad quality (see
mp3). 
But, of course, this way you have no guarantee about the audio
quality.

> speech codecs are standard now.  Otherwise, it is
difficult for customers to
> know what they are getting.  It also makes it difficult for telecom vendors
to
> provide reasonable assurance on indemnification issues.

Do you mean the audio quality? How to get an objective measure
of quality (for the court case)?

BTW, which codec do you mean by "other than mp3"? For all MPEG
Encoders (mp2, mp3, aac, etc.) only the bitstream is defined,
and some newer codecs like OggVorbis do it in the same way. I
don't know exactly, but I think even the proprietary codecs
like ATRAC, AC-3, DTS, etc. do not specify a bit-exact
description of the encoded bitstream.

Regards,
Alexander Lerch

-- 
alexander lerch
lerch@lerc...

zplane.development
holsteinische str. 39-42
12161 berlin
 fon: +49.30.854 09 15.0
 fax: +49.30.854 09 15.5
	
Mihir Narendra Mody-

>  You can go to www.mpeg.org and its gives many
links to MP3 papers,
>software etc. You can buy standard from ISO site.

One problem we have with calling the ISO site "standard":  there are
no standard 
input and output vectors to use for bit-exactness verification.  I.e. there are 
no given .wav (or .aif) input files, .mp3 output files, and resulting decoded 
.wav files for comparison.

We also tried to get comparison files from Fraunhofer, even going so far as to 
contact upper management.  Nothing.  Fraunhofer has some .mp3 files on their 
site, but they appear to withhold the original .wav files.

So everyone's implementation of MP3 is possibly -- likely at least somewhat -- 
different, as long as it sounds Ok?  I guess so.

Any comments on this issue are greatly appreciated.

Jeff Brower
DSP sw/hw engineer
Signalogic
	
Mihir Narendra Mody-

Thanks for your informative reply.  It's more than I thought was there, but 
still too imprecise.  There should also be conformance streams for the encoder, 
given certain parameters.

Given different levels of constraint, it should be possible to perform bit-exact

verification testing all the way through encode and decode -- just like standard

speech codecs.  I have a feeling this is one reason that audio codecs other than

MP3 may become the standard in AoIP / AoP systems in the future, the way certain

speech codecs are standard now.  Otherwise, it is difficult for customers to 
know what they are getting.  It also makes it difficult for telecom vendors to 
provide reasonable assurance on indemnification issues.

Jeff Brower
Signalogic
	>HI Jeff,
>   When you buy CD from standard, in CD they give
some conformance
>streams. Your decoder should able to decode and final output error
>should not exceed the error specified from reference file. The exact
>procedure for testing of decoder using mp3 file is specified in part4
>of standard (conformance testing).
>  For encoder only restriction is that the bitstream should confirm
>the standard. The quality of encoded file depends on Encoder
>implementation ( quantizor, psycho-acoustics etc). So output encoded
>files of MP3 encoder are of different audio quality depending on
>the encoder implemetation, But decoder will able to decode all
>mp3 files. For high quality encoder, you have to pay more.
>Regards
>Mihir Mody
>Saken Communication technology Ltd
>
>
>
>On Mon, 19 Feb 2001, Jeff Brower wrote:
>
>> Mihir Narendra Mody-
>>
>> >  You can go to www.mpeg.org and its gives many links to MP3
papers,
>> >software etc. You can buy standard from ISO site.
>>
>> One problem we have with calling the ISO site "standard": 
there are no 
standard
>> input and output vectors to use for
bit-exactness verification.  I.e. there 
are
>> no given .wav (or .aif) input files, .mp3
output files, and resulting decoded
>> .wav files for comparison.
>>
>> We also tried to get comparison files from Fraunhofer, even going so
far as 
to
>> contact upper management.  Nothing. 
Fraunhofer has some .mp3 files on their
>> site, but they appear to withhold the original .wav files.
>>
>> So everyone's implementation of MP3 is possibly -- likely at least
somewhat 
--
>> different, as long as it sounds Ok?  I guess
so.
>>
>> Any comments on this issue are greatly appreciated.
>>
>> Jeff Brower
>> DSP sw/hw engineer
>> Signalogic
	
Hi Jeff,
   In Mp3 standard, the decoder is standard and no much space is left
for differernt manufacture to change it. Only they can use fast algorithm
and resulting some error should not exceed specified value in standard.
   The real thing or chanllenge they kept on encoder. The algorithm
is not standardised but bitstream is standardized. So differnt manufacture
can come up with differnt implementation and quality of encoding is
differnt. The more knowlege and experiments done by manufactures will
give better quality. For better quality you have to pay more. There are
some reference wave files (like castanet, germal male speech) are there
in audio coding. All manufacture have to specify the quality (MOS score)
of this refence file by their encoder. There are some special softeare
also
avaible (like OPERA from OPTICOM) which will give quality of encoded
files. So while buuying you can ask the qulity of encoder on this
benchmark signals.
 I hope this will clarify your doubt
Regards
Mihir N Mody
Sasken Communication Techology Ltd
Bangalore, INDIA
	On Mon, 19 Feb 2001, Jeff Brower wrote:

> Mihir Narendra Mody-
>
> Thanks for your informative reply.  It's more than I thought was there, but
> still too imprecise.  There should also be conformance streams for the
encoder,
> given certain parameters.
>
> Given different levels of constraint, it should be possible to perform
bit-exact
> verification testing all the way through encode and decode -- just like
standard
> speech codecs.  I have a feeling this is one reason that audio codecs other
than
> MP3 may become the standard in AoIP / AoP systems in the future, the way
certain
> speech codecs are standard now.  Otherwise, it is difficult for customers
to
> know what they are getting.  It also makes it difficult for telecom vendors
to
> provide reasonable assurance on indemnification issues.
>
> Jeff Brower
> Signalogic
>
>
> >HI Jeff,
> >   When you buy CD from standard, in CD they give some conformance
> >streams. Your decoder should able to decode and final output error
> >should not exceed the error specified from reference file. The exact
> >procedure for testing of decoder using mp3 file is specified in part4
> >of standard (conformance testing).
> >  For encoder only restriction is that the bitstream should confirm
> >the standard. The quality of encoded file depends on Encoder
> >implementation ( quantizor, psycho-acoustics etc). So output encoded
> >files of MP3 encoder are of different audio quality depending on
> >the encoder implemetation, But decoder will able to decode all
> >mp3 files. For high quality encoder, you have to pay more.
> >Regards
> >Mihir Mody
> >Saken Communication technology Ltd
> >
> >
> >
> >On Mon, 19 Feb 2001, Jeff Brower wrote:
> >
> >> Mihir Narendra Mody-
> >>
> >> >  You can go to www.mpeg.org and its gives many links to MP3
papers,
> >> >software etc. You can buy standard from ISO site.
> >>
> >> One problem we have with calling the ISO site
"standard":  there are no
> standard
> >> input and output vectors to use for bit-exactness verification. 
I.e. there
> are
> >> no given .wav (or .aif) input files, .mp3 output files, and
resulting decoded
> >> .wav files for comparison.
> >>
> >> We also tried to get comparison files from Fraunhofer, even going
so far as
> to
> >> contact upper management.  Nothing.  Fraunhofer has some .mp3
files on their
> >> site, but they appear to withhold the original .wav files.
> >>
> >> So everyone's implementation of MP3 is possibly -- likely at least
somewhat
> --
> >> different, as long as it sounds Ok?  I guess so.
> >>
> >> Any comments on this issue are greatly appreciated.
> >>
> >> Jeff Brower
> >> DSP sw/hw engineer
> >> Signalogic
>
	
Hello Mihir Narendra Mody,

Mihir Narendra Mody wrote:
> 
> There are some special softeare
> also
> avaible (like OPERA from OPTICOM) which will give quality of encoded
> files. So while buuying you can ask the qulity of encoder on this
> benchmark signals.

I spent some time with implementing PEAQ (ITU-R BS.1387),
which is the application in OPERA to rate high quality codecs,
and I can say it will not supersede listening tests. The
correlation of PEAQ ratings with results of listening tests is
not very high. 
Also, PEAQ is only designed for high quality audio codecs, and
codecs with middle quality can not be evaluated. For low
quality there are only some speech quality measurement tools
(PSQM is integrated in OPERA, I think).
In particular applications, objective measurement tools make
sense, but the rating of some encoded-decoded audio files will
not be very useful for quality evaluation of the codec.

Regards,
Alexander

-- 
alexander lerch
lerch@lerc...

zplane.development
holsteinische str. 39-42
12161 berlin
 fon: +49.30.854 09 15.0
 fax: +49.30.854 09 15.5
	
Mihir Narendra Mody-

>So differnt manufacture
>can come up with differnt implementation and quality of encoding is
>differnt. The more knowlege and experiments done by manufactures will
>give better quality.

So where is the baseline?  Where is the minimum level, testable by an 
independently sponsored and impartial testing organization?  Such a test must be

developed for MP3 (and/or other audio codecs) before they can be used in telecom

and carrier class products.  The closest telecom analogy seems to be echo 
cancellers, which also are improved constantly by different manufacturers.  But 
-- for echo cancellers there is a widely accepted performance standard (G.165, 
G.168).  I guess MP3 is just not there yet.

Thanks very much for discussion.  I think I understand the problem better now.

Jeff Brower
DSP sw/hw engineer
Signalogic
	On Tue, 20 Feb 2001, Mihir Narendra Mody <mihir@mihi...> wrote:
>Hi Jeff,
>   In Mp3 standard, the decoder is standard and no much space is left
>for differernt manufacture to change it. Only they can use fast algorithm
>and resulting some error should not exceed specified value in standard.
>   The real thing or chanllenge they kept on encoder. The algorithm
>is not standardised but bitstream is standardized. So differnt manufacture
>can come up with differnt implementation and quality of encoding is
>differnt. The more knowlege and experiments done by manufactures will
>give better quality. For better quality you have to pay more. There are
>some reference wave files (like castanet, germal male speech) are there
>in audio coding. All manufacture have to specify the quality (MOS score)
>of this refence file by their encoder. There are some special softeare
>also
>avaible (like OPERA from OPTICOM) which will give quality of encoded
>files. So while buuying you can ask the qulity of encoder on this
>benchmark signals.
> I hope this will clarify your doubt
>Regards
>Mihir N Mody
>Sasken Communication Techology Ltd
>Bangalore, INDIA
>
>
>
>On Mon, 19 Feb 2001, Jeff Brower wrote:
>
>> Mihir Narendra Mody-
>>
>> Thanks for your informative reply.  It's more than I thought was there,
but
>> still too imprecise.  There should also be conformance streams for the 
encoder,
>> given certain parameters.
>>
>> Given different levels of constraint, it should be possible to perform 
bit-exact
>> verification testing all the way through
encode and decode -- just like 
standard
>> speech codecs.  I have a feeling this is one
reason that audio codecs other 
than
>> MP3 may become the standard in AoIP / AoP
systems in the future, the way 
certain
>> speech codecs are standard now.  Otherwise, it
is difficult for customers to
>> know what they are getting.  It also makes it difficult for telecom
vendors 
to
>> provide reasonable assurance on
indemnification issues.
>>
>> Jeff Brower
>> Signalogic
>>
>>
>> >HI Jeff,
>> >   When you buy CD from standard, in CD they give some conformance
>> >streams. Your decoder should able to decode and final output error
>> >should not exceed the error specified from reference file. The
exact
>> >procedure for testing of decoder using mp3 file is specified in
part4
>> >of standard (conformance testing).
>> >  For encoder only restriction is that the bitstream should confirm
>> >the standard. The quality of encoded file depends on Encoder
>> >implementation ( quantizor, psycho-acoustics etc). So output
encoded
>> >files of MP3 encoder are of different audio quality depending on
>> >the encoder implemetation, But decoder will able to decode all
>> >mp3 files. For high quality encoder, you have to pay more.
>> >Regards
>> >Mihir Mody
>> >Saken Communication technology Ltd
>> >
>> >
>> >
>> >On Mon, 19 Feb 2001, Jeff Brower wrote:
>> >
>> >> Mihir Narendra Mody-
>> >>
>> >> >  You can go to www.mpeg.org and its gives many links to
MP3 papers,
>> >> >software etc. You can buy standard from ISO site.
>> >>
>> >> One problem we have with calling the ISO site
"standard":  there are no
>> standard
>> >> input and output vectors to use for bit-exactness
verification.  I.e. 
there
>> are
>> >> no given .wav (or .aif) input files, .mp3 output files, and
resulting 
decoded
>> >> .wav files for comparison.
>> >>
>> >> We also tried to get comparison files from Fraunhofer, even
going so far 
as
>> to
>> >> contact upper management.  Nothing.  Fraunhofer has some .mp3
files on 
their
>> >> site, but they appear to withhold the
original .wav files.
>> >>
>> >> So everyone's implementation of MP3 is possibly -- likely at
least 
somewhat
>> --
>> >> different, as long as it sounds Ok?  I guess so.
>> >>
>> >> Any comments on this issue are greatly appreciated.
>> >>
>> >> Jeff Brower
>> >> DSP sw/hw engineer
>> >> Signalogic