naturerecordists
[Top] [All Lists]

Re: ATRAC don't get no respect

Subject: Re: ATRAC don't get no respect
From: Walter Knapp <>
Date: Fri, 25 Apr 2003 14:22:51 -0400
Raimund Specht wrote:
> --- In  Walter Knapp <> wrot=
e:
>
>
>>Next, examine exactly the gain used on the mp3 and original. This
>
> has a
>
>>very strong bearing on the resulting display.
>
>
> I did not change the gain between the two examples. Also, the
> spectrogram contrast was not changed.=20=20

However it was a simple matter to measure that. Overall your mp3 sample
is boosted about 1dB over the original. Not uncommon for mp3 encoders to
give a different level slightly. I did not dissect it on a sample by
sample basis, probably even greater errors in that. Visually I could see
the differences in the overall waveform display switching back and forth
between the two samples.

I should note I have 7 years experience using ATRAC compression for
scientific work. There I have constantly monitored it in detail. I've
not treated mp3 to the same level of examination on any consistent
basis. I can hear it doing things ATRAC does not do. I mostly use it for
it's original purpose, making sound files small enough to send over the
internet. And mostly encode by listening, moving to more and more
compression until I hear things I don't want, then backing off until
they are gone. Very crude compared to what I do with my main recordings.

>>It's a test of how well mp3 handles sharp high frequency transients,
>>waveforms that closely resemble clipping artifacts and so on. Looks
>
> a
>
>>lot like a square wave generator was used to generate a lot of the
>>original. To assume this bears any resemblance to natural sound is
>>interesting in what it tells about people.
>
>
> Such sharp transients are often present in animal sounds (e.g. some
> song birds or grasshoppers). There is no clippng involved! At least
> not at the upper edge of the dynamic range. But there is of course
> some kind of clipping at the bottom edge of the dynamic range. This
> is the 'magic' behind the both the ATRAC and MP3 systems. Hopefully,
> would will understand this correctly.

Look at it expanded out enough too look at the waveform of the original
you will find quite a few notched peaks, a common form of clipping
artifact. I know it's not true clipping, but has the same waveform.
Sharply notched peaks are virtually a unknown in natural recordings. And
I have looked at the sharp transients of animals of all types. The ones
I'm talking about are single sample notches and are very common in this
original. You also have sample variations throughout where it's runing a
intersample transition from full minus signal to full plus. Real natural
environments don't have that kind of high frequency content.

> The original test file was generated using a flexible sine generator
> (and noise generator). That special software can also be used to
> generate very naturalistic animal sounds. Download the demo version
> of Avisoft-SASLab Pro from:
>
> www.avisoft.info/downloa_.htm (first item)

If this was a example of it's naturalistic sound, no thanks. I've never
seen such a consistently crummy waveform out of a nature recording, even
those with clipping. Like sonograms, I also look at the waveforms of my
recordings. The most common thing for me to find when I hear some error
in a recording is a few wavetops with clipping. Even crude redraws of
those will generally remove the problem. Tedious work, so I try to avoid
the need. But, for a rare recording I'll do it.

> Go to Edit/Synthesizer/"Synthesizer(graphically)" and select
> File/Load and press the play button. You will hear a synthesized
> song of the European Nightingale. Also use the mouse to modify some
> of the elements and play it back again. Unfortunately, there is no
> MAC version, but you could use the Virtual PC software to emulate
> Windows.

I have a pc here as well. But, as I said this sure did not look like it
was doing anything I had a need for. There are mac signal generators
that generate good signals.

> I used these very artificial sound structures to make the effects
> more visible on a spectrogram. I could also prepare some more
> natural sounds. I will do this soon. However, I'm afraid that you
> will still find many additional hair-splitting arguments to
> disrespect my demonstrations It seems to be some kind of religious
> believe, that lead to that ignorance (I'm very sorry for this strong
> comment). I should therefore better respect this religious minority
> and break off any further public criticism...
>
> Please also note, that I'm not one of those fat and neglected
> computer freaks, who never leave the laboratory. I do practise
> animal sound recording in the field too (for those who care about
> that)!

I figured you might do so, or else you were really lost. Try the same
analysis on real outdoor recordings you have made. This is nature
recording, not a course in seeing if we can invent some exotic sound to
trip up compression. That properly should be for rec.audio.pro. They
spend most of their time on that sort of trivia. For a practical person
very boring.

My position all along has been that ATRAC is good enough for the job at
hand. That includes nearly all science recording. I do scientific
recording. If you remember, we were talking about running fingers across
chalkboards, cutting styrofoam. I'd record the chalkboard if I had one.
Slate chalkboards have a better upsetting effect than the painted ones.
I had a 4x8' slate one when I was growing up, probably still at my
parents house. I know that sound in great detail. I also know that many
people with very poor hearing will react to that sound, which means it's
not that complex, or high frequency.

I should note it's fairly trivial to filter out those artifacts you are
so worried about. Use a dynamic range filter on most of it, should work.
Anyway, looks removable that way. In nature recording it would just be
part of the background noise people were filtering. And a minor part at
that as it's not audible.

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