naturerecordists
[Top] [All Lists]

Re: ATRAC don't get no respect

Subject: Re: ATRAC don't get no respect
From: "Raimund Specht" <>
Date: Thu, 24 Apr 2003 11:08:07 -0000
--- In  Walter Knapp <> 
wrote:

>I've been waiting for 7 years for someone to actually show these 
effects
>that are so far just imaginary, nothing more than a hypothesis. I 
have
>done considerable looking on my own as well, I'm a scientist. It's 
time
>someone put up or shut up. All we have had is Cornell trying to 
pass off
>digital clipping as the effect of ATRAC. Still trying to pass it 
off as
>the effect of ATRAC today on their website. That's how frantic some 
are
>to discredit ATRAC. I don't understand this sillyness, it's 
definitely
>not scientific. A scientist knows that his hypothesis must be 
tested or
>it's nothing. A scientist provides real data and shows how that's 
relevant.

>So, do it, show us, and demonstrate how it makes a difference to the
>animals. I've done every resolution of sonogram available to me 
without
>finding what you talk of. ...

Walt, I have just prepared a quick comparison of both uncompressed 
and compressed digital audio as you suggested:

http://www.avisoft-saslab.com/compression/compression.htm

I guess, that you will argue, that these tests were made with MP3 
and not with the latest ATRAC system. However, I will NOT buy one of 
these expensive MiniDisk machines. The underlaying principles are 
always the same. The improvements may lead to better listening 
pleasure, but the loss of data will remain.

>ATRAC is very, very predictable. It's a fixed circuit on a chip, 
will
>respond exactly the same way every time. If you want unpredictable, 
go
>to DAT tape, which is considered to be so perfect. I went to MD for 
it's
>predictability.

Not true! There seem to be dramatic differences between the various 
versions of ATRAC. Each version may encode differently. This is 
because the ATRAC system does not demand a unified compression 
regime. It is up to the manufacturer to implement different psycho-
acoustic models. Also, the more sophisticated versions as MP3 with 
its 'bit reservoir' feature will add even more irregular effects.

>This is where your inexperience with ATRAC shows. ATRAC saves none 
of
>the original information. Understand that exactly, so I'll repeat 
it,
>ATRAC saves none of the original information. What it saves amounts 
to
>instructions to a synthesizer. This is one of the big errors made in
>understanding ATRAC, making the simplistic assumption that it 
tosses 4
>out of 5 samples. If that's all it did, no matter how 
intelligently, it
>would not be likely to do very well. As we know from midi extremely
>small amounts of data fed to a synthesizer can produce extremely 
complex
>sounds.

I'm sorry, but I would wish, that you improved your reading skills...

I did not say that there would be a simple sample rate reduction of 
5:1 !!! But of course, that would be one way to reduce the data 
rate. For instance, you could use 8 bit instead of 16 bit and a 
sample rate of 22.05 kHz instead of 44.1 kHz, That would provide a 
data reduction of 4:1 and the results might be not too ugly for 
noisy nature recordings. 

There is no pure synthesizer in ATRAC. The time-domain data are 
transformed into the frequency domain and the resolution of these 
data (which correspond to the time domain data) is reduced in order 
to shrink the data stream. These frequency domain representation of 
the signal is saved onto the disk. The decoder transforms these data 
back into the time-domain (using DCT, something similar to FFT).  

I say it AGAIN, I do NOT argue, that ATRAC or any other compression 
technology is bad for the purpose it was designed for (human 
listening). But you should know the limitations (we already leard 
this on parabolas...) when you use that technology for research in 
animal sound communication.

>It's also unwise to think that ATRAC chooses which bit depth to 
reduce
>based on human auditory systems. Nearly all of that it does is 
based on
>things like waveform shapes, not the much brought out human 
hearing. It
>chooses bit depths based on how easily the shape can be described. 
It
>assigns more data to more complex parts. Will it limit out doing 
this?
>Yes it should eventually. But long after CD, which has the same 
sampling
>rate, but applies it evenly no matter if needed or not. Or DAT, 
which
>also is dumb sampling, no choosing. Both of those waste a lot of the
>bandwidth they have.

That's definitely false. Follow the links to the description of 
ATRAC I submitted on my website and you will see what is really 
going on. You are mixing some truth with a lot of fairytales. I'm an 
engineer and have some experience with digital signal processing. 
Even if I have not used ATRAC much, I can explain, what's going on 
(after reading the technical specifications).

I'm very sorry to say all this so clearly, but I just want to 
prevent that others become confused about the truth on data 
compression technology.  

I'm a little bit tired of discussing these things in that way. Make 
your own experiments with a clean test signal and you will 
understand what I mean.

Regards,
Raimund



________________________________________________________________________
________________________________________________________________________

<Prev in Thread] Current Thread [Next in Thread>
Admin

The University of NSW School of Computer and Engineering takes no responsibility for the contents of this archive. It is purely a compilation of material sent by many people to the naturerecordists mailing list. It has not been checked for accuracy nor its content verified in any way. If you wish to get material removed from the archive or have other queries about the archive e-mail Andrew Taylor at this address: andrewt@cse.unsw.EDU.AU