MP3 coding (was Current trends in audio)
 

Johan Helsingius said:
On 22-01-17 13:57, Richard Robinson wrote:

If you'd excuse a thread-swerve, while the subject's mp3 - I'm not familiar
with the internals of mp3, and there's a thing I've never been sure of.
Namely, what happens if you do the compression twice ? I mean, if you
convert an mp3 file back into something lossless, then recode that as mp3,
what gets lost on the second compression ? Does it only throw away what was
already thrown away on the first compression (ie the quality of the second
generation isn't much worse than the first), or do you lose the same amount
of 'quality' all over again ? (I'm guessing somewhere between the 2, but I
have nothing much to go on except masculine intuition).

Yes, "somewhere between the 2" is pretty much the same answer.

Double compression degrades the result a bit further, but not much
as long as you use the same mp3 encoder and the same bit rate. Changing
encoder and/or bit rate might result in different decisions about
what to discard, thus degrading the result further.

While still not understanding the coding details, that seems to make good
sense, yes. Thanks.


--
Richard Robinson
"The whole plan hinged upon the natural curiosity of potatoes" - S. Lem

My email address is at http://www.qualmograph.org.uk/contact.html

Current trends in audio
 

On 23-01-17 04:28, Phil Allison wrote:

** Yes, and the result shows how useless all A then B tests are and by implication variations like ABX.

You seem to keep assuming that ABX tests don't allow for instantaneous
back-and-forth switching, but even the original ABX Company paper
(by David Clark) from 1982 talks about their device being equipped
with user-controllable switchover pushbuttons (and discusses ways
to avoid the relay switchover being audible).

http://www.aes.org/e-lib/browse.cfm?elib=3839

Julf

MP3 coding (was Current trends in audio)
 

Jim Lesurf said:
In article , Johan Helsingius
wrote:
On 22-01-17 13:57, Richard Robinson wrote:

If you'd excuse a thread-swerve, while the subject's mp3 - I'm not
familiar with the internals of mp3, and there's a thing I've never
been sure of. Namely, what happens if you do the compression twice ? I
mean, if you convert an mp3 file back into something lossless, then
recode that as mp3, what gets lost on the second compression ? Does it
only throw away what was already thrown away on the first compression
(ie the quality of the second generation isn't much worse than the
first), or do you lose the same amount of 'quality' all over again ?
(I'm guessing somewhere between the 2, but I have nothing much to go
on except masculine intuition).

Yes, "somewhere between the 2" is pretty much the same answer.

Double compression degrades the result a bit further, but not much as
long as you use the same mp3 encoder and the same bit rate. Changing
encoder and/or bit rate might result in different decisions about what
to discard, thus degrading the result further.

Anyone interested can probably do an analogous experiment by repeatedly
jpeg encoding and rendering an image though a series of sucessive cycles.
The image will tend to degrade after each cycle. But the rate at which it
does so will vary with factors like the amount of data reduction used and
the 'rules' being employed. Some encoders may be better than others.

That's a thing that hadn't occured to me. Is it a similiar algorithm ? It
would be an easier & less time-consuming test.

Thus it becomes obvious that for any 'chain' where end-quality matters
you'd avoid data reduction (aka lossy compression) until the last lap!

Well, yes, I'd say it always was. But.

I ripped my dying C90s before I could afford the storage for lossless, so
they're all mp3s and nothing to be done about it. And I did a reasonable
job, but there was an awful lot to do and sometimes I slipped up; so I
listen to something and notice little glitches that maybe should be tidied
up. And then I start wondering how much further degradation that would
entail ... and don't seem to find time for a proper "suck it and see", or to
read (understand) the code in question.

Mind you (thinks: "duh!"), if I was to get round to it now I wouldn't recode
back to MP3, so maybe it's only of academic interest. But still kind of
interesting.

--
Richard Robinson
"The whole plan hinged upon the natural curiosity of potatoes" - S. Lem

My email address is at http://www.qualmograph.org.uk/contact.html

MP3 coding (was Current trends in audio)
 

On 22/01/2017 14:06, Johan Helsingius wrote:
On 22-01-17 13:57, Richard Robinson wrote:

If you'd excuse a thread-swerve, while the subject's mp3 - I'm not familiar
with the internals of mp3, and there's a thing I've never been sure of.
Namely, what happens if you do the compression twice ? I mean, if you
convert an mp3 file back into something lossless, then recode that as mp3,
what gets lost on the second compression ? Does it only throw away what was
already thrown away on the first compression (ie the quality of the second
generation isn't much worse than the first), or do you lose the same amount
of 'quality' all over again ? (I'm guessing somewhere between the 2, but I
have nothing much to go on except masculine intuition).

Yes, "somewhere between the 2" is pretty much the same answer.

Double compression degrades the result a bit further, but not much
as long as you use the same mp3 encoder and the same bit rate. Changing
encoder and/or bit rate might result in different decisions about
what to discard, thus degrading the result further.


But not, IIUC, if you use lossless encoding like flac or Apple lossless?
You can go wav-flac-wav endlessly and always have the same wav?

Still, I am surprised that mp3s don't have metadata to enable a
successful rebuild to the original.
--
Cheers, Rob

Current trends in audio
 

On 23/01/2017 09:36, Jim Lesurf wrote:
In article , Phil
Allison wrote:

I daresay a variant of it would work well with two speakers.

** Speakers DO sound different, as one would expect, since they do not
test the same. Having an instant changeover makes comparison tests very
easy though and eliminates imaginary differences.

A potential problem here would be that speakers, in general, don't stop
radiating the instant you abruptly cut off input to them. So when you
switched, one would still be 'ringing down' in its own way as the other
speaker 'gets going'.

Unless you have a one-note hi-Q subwoofer this isn't a problem.

--
Eiron.

Current trends in audio
 

Johan Helsingius wrote:

Phil Allison wrote:

** Yes, and the result shows how useless all A then B tests are
and by implication variations like ABX.


You seem to keep assuming that ABX tests don't allow for instantaneous
back-and-forth switching,


** According the Arny Kruegar who built the relay boxes - they never did in practice.


but even the original ABX Company paper
(by David Clark) from 1982 talks about their device being equipped
with user-controllable switchover pushbuttons (and discusses ways
to avoid the relay switchover being audible).

http://www.aes.org/e-lib/browse.cfm?elib=3839


** You gotta be ****ing kidding us !!!!!!!!!!

Pay $33 to read complete tripe ???

**** YOU !!!

FYI:

I have tolerated this pig ignorant, Dutch MP3 fascist far too long.

From now on, no more Mr Nice Guy.

Hey Joahan, get your hands off it.

This NG is about home Hi-Fi and you have no ****ing clue what that even is.

You have no idea what anyone thinks - except you.

Cos you are a boorish, autistic moron.

Just like all Dutch pigs.




..... Phil

MP3 coding (was Current trends in audio)
 

RJH said:
On 22/01/2017 14:06, Johan Helsingius wrote:
On 22-01-17 13:57, Richard Robinson wrote:

If you'd excuse a thread-swerve, while the subject's mp3 - I'm not familiar
with the internals of mp3, and there's a thing I've never been sure of.
Namely, what happens if you do the compression twice ? I mean, if you
convert an mp3 file back into something lossless, then recode that as mp3,
what gets lost on the second compression ? Does it only throw away what was
already thrown away on the first compression (ie the quality of the second
generation isn't much worse than the first), or do you lose the same amount
of 'quality' all over again ? (I'm guessing somewhere between the 2, but I
have nothing much to go on except masculine intuition).

Yes, "somewhere between the 2" is pretty much the same answer.

Double compression degrades the result a bit further, but not much
as long as you use the same mp3 encoder and the same bit rate. Changing
encoder and/or bit rate might result in different decisions about
what to discard, thus degrading the result further.


But not, IIUC, if you use lossless encoding like flac or Apple lossless?
You can go wav-flac-wav endlessly and always have the same wav?

Yes.

Still, I am surprised that mp3s don't have metadata to enable a
successful rebuild to the original.

Could that be done, without storing the original data ?

--
Richard Robinson
"The whole plan hinged upon the natural curiosity of potatoes" - S. Lem

My email address is at http://www.qualmograph.org.uk/contact.html

MP3 coding (was Current trends in audio)
 

In article ,
Richard
Robinson wrote:

Still, I am surprised that mp3s don't have metadata to enable a
successful rebuild to the original.

Could that be done, without storing the original data ?

The 'physical' measure/unit of the quantity of information is 'bits'. It
follows from the basics of Information Theory that a given 'message' will
contain some amount of information. You can transform or shuffle this to
try and remove waste, but then end up with the result not being any smaller
if you don't want to lose any information.

Attempts to do otherwise either:

1) Throw away some info and hope you won't notice.

and/or

2) Cheat. :-) By storing the info somewhere else and using what you get as
the message an instruction to use that. Magic of Indirection. :-)

Either way, the info has to be somewhere, and accessible to you.

For music, LPCM tends to be wasteful. So you can reshuffle how it is
represented and that tends to reduce the number of bits required. Flac
works pretty well for this. Although that has the snag that genuine random
background noise is preserved just as if it were 'real' information. This
is the main reason many 'High Rez' flac files are so big. Noise, carefully
preserved in every detail, not actually shedloads of more *musical* info.

MP3 (and other lossy systems) tend to throw away info in the hope you won't
care or notice. You can do this to *some* extent if done with due care and
not affect the result noticably. But inevitably, someone then wants to wind
up the compression to '11'... 8-

Jim

--
Please use the address on the audiomisc page if you wish to email me.
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc http://www.audiomisc.co.uk/index.html

Current trends in audio
 

In article ,
Graeme Wall wrote:
Not quite sure what you mean. The OnDigital system effectively morphed
into FreeView.

When the BBC got involved.

Quite. The original On Digital setup was a consortium of Thatcher's
favourite ITV companies whose only motive was to make a lot of money. But
sadly hadn't a hope when up against the likes of the dirty digger.


I had OnDigital from the start, and didn't have problems with picture
break up. But like all digital systems it reacts differently to a poor
signal than analogue.


AIUI the initial OnDigital service was not marketed to home subscribers,
but I could be wrong!

Think it always was - the idea being they could charge a subscription for
it. As well as things like footie for pubs. Rather like Sky is now. I only
wanted it for HD, though.

There were problems initially with low field strengths which continued
until analogue was switched off. Not a problem here - I can see the CP
mast from the rear of the house.

--
*Change is inevitable, except from a vending machine.

Dave Plowman London SW
To e-mail, change noise into sound.

MP3 coding (was Current trends in audio)
 

In article ,
Richard
Robinson wrote:

That's a thing that hadn't occured to me. Is it a similiar algorithm ?
It would be an easier & less time-consuming test.

At a the simplest mathematical level, yes. MP3 and JPEG use the same basis.

Each uses a Fourier Transform to generate a series of 'spectra'. These then
have the values which the encoder examines and decides which ones you won't
mind being left out when the results are stored. (Some results are stored,
but possibly with lower resolution which means they take fewer bits.)

The differences are that MP3 does a 1D temporal spectrum per chunk/channel
whereas JPEG uses 2D spatial ones. But from an basic IT POV that's a mere
detail.

They key here is actually not just the output bitrate but what are called
"judgement rules". These are some instructions given to the encoders to tel
them how to decide what to omit when encoding, and what to keep, but with
some level or reduction in resolution. Optimising these gets the best
trade-off between the size of the result and its quality. But you can't get
owt for nowt, hence comments I've made in another posting. :-)

The decoder then runs this process backwards, but of course doesn't have
the 'lost' values or the original resolution for all the others. So can't
restore all the original details.

In an ideal world, using the same encoder and decoded again and again might
mean the accumulated losses don't grow because each time it finds a
'perfect match'. But in reality I'm not sure I'd expect this from real
encoders. And if you made any changes to the details between a decode and
an encode, it's a cast iron certainty that the next encode would lose
something.

Jim

--
Please use the address on the audiomisc page if you wish to email me.
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc http://www.audiomisc.co.uk/index.html