naturerecordists
[Top] [All Lists]

Re: Only a question

Subject: Re: Only a question
From: Walter Knapp <>
Date: Tue, 20 Jul 2004 16:24:00 -0400
From: jan.larsson <>

>
>
> As you say. There is no denying that ATRAC can do a  very good job. And
> there is no idea in denying that the job of ATRAC (and MP3 and AAC and
> AC3 ....) involves removing stuff in order to lower the bit-rate. And
> it is a good deal of stuff that gets removed.
>
> If you are unsecure about this then do your own test very simply.
>
> Use a couple of different tracks of material recorded with PCM. Some
> soundscape, some birds, some frogs. Make a digital ATRAC-copy of each.
>
> Then make a digital transfer back from ATRAC to PCM (a minidisc with
> digital output).
>
> Now you make a CDR with three cd-tracks per original track. The first
> is always the original as reference. The other two are the original and
> the ATRAC always in random order.
>
> Now you can distribute the CDR and people can listen on their own
> well-known reference system and try to pick the ATRAC.
>
> This is very easy to setup. You could even do it as a list exercise as
> long as people agree not to talk about it until everybody has reported
> their results (that would be the hard part ;=3D)).

Old history, been done many times. Even been done in this group.

I managed to work out a pathway that just tested ATRAC. That's not near
as simple as it sounds, most such tests test ATRAC and the rest of the
recorder, quite a different thing. I took advantage of the exact sincere
copying from CD to MD in a Sony MXD-D3 (which is ATRAC 4, a older
version) to get around the rest.

I provided three part samples, of nature recording subjects, in which a
clip was repeated three times. These were random ATRAC or no ATRAC. The
group voted on their identifications by listening any way they wished to
listen, and were then allowed to analyze it any way they wanted and
provide a 2nd vote for that. For those without sonogram capability I
provided sonograms that people could consult for the 2nd part. Votes
were not posted on the group and people were not to compare notes. On a
specified date I tallied the votes received and reported the results.

Results, by listening those that participated had very close to a 50%
error rate, about what's expected if just flipping a coin to decide.
Surprisingly, the correct percentage only rose to 67% when using any
analysis they wanted. Including my sonograms, which did show the
difference that was common for ATRAC 4, very slight decrease in
background noise at frequencies above 16 kHz. These were not associated
with any call. They were sounds that were more than 50 dB below the
other sounds found in that frequency range.

Used to be that test (with the key) was up in our binary area on Doug's
site, but it appears it's been removed. Since it was older ATRAC, that's
probably appropriate.

This result is fairly typical for such tests. ATRAC for some time has
been capable of passing that test easily. And it's been improved since
then. ATRAC 4.5 does not show near as much decrease in the very faint
sounds at high frequencies. You have to go above 18 kHz to find any of
it. As far as I'm concerned if that high level is critical you should be
using higher sampling rates, nothing doing 44kHz sampling rates does a
good job that close to the limit.

One of the most amusing things about this sort of test to me is how the
"golden ears" crowd run from participating. They are very creative about
reasons why their real capabilities should not be tested.

Walt






________________________________________________________________________
________________________________________________________________________

<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