SonicStage 3.2: ATRAC3plus Frequency Analysis

[Ishiyoshi]

ATRAC3plus Frequency Analysis

To complement the subjective listening tests done on the new bit rates of ATRAC3plus --> notably 320kbps (in this thread), I have conducted a frequency analysis for your technical curiosity.

*Thanks to greenmachine for the concise Adobe Audition guide.

I. A frequency analysis between ATRAC3plus 320kbps, 256kbps and 192kbps.

Original file: Nina Simone's I Loves Porgy.

Green: original uncompressed file

Red: ATRAC3plus, 320kbps

Blue: ATRAC3plus, 256kbps

Yellow: ATRAC3plus, 192kbps

Linear view

Logarithmic view

II. A frequency analysis between ATRAC3plus 160kbps, 128kbps, 96kbps, 64kbps and 48kbps.

Original file: Nina Simone's I Loves Porgy.

Green: original uncompressed file

Red: ATRAC3plus, 160kbps

Blue: ATRAC3plus, 128kbps

Linear View

Original file: Nina Simone's I Loves Porgy.

Green: original uncompressed file

Red: ATRAC3plus, 96kbps

Blue: ATRAC3plus, 64kbps

Yellow: ATRAC3plus, 48kbps

Linear View

As greenmachine illustrated in his earlier frequency analysis, there are no significant difference between ATRAC3plus 320kbps and 256kbps. They are almost identical in terms of quality as shown in the graph above. However, one can easily discern ATRAC3plus, 192kbps from 320kbps and 256kbps.

Discuss accordingly.

[greenmachine]

As i've said before, the frequency analysis alone doesn't tell the whole truth about sound quality, it's just a rough orientation point to see if there's happening something exceptionally weird, like with the native mp3 playback for 2nd gen md portables.

All you can see here is that a lowpass filter at 18kHz is used for 192 kbps, which is needed for effective compression - you have to draw the line somewhere, after all it's a lossy compression method.

I recently did some tests with lowpass filters in various musical pieces. I was hard pressed to hear a filter above about 16kHz, above 17kHz it was next to impossible to detect for me. Although this value may vary between users, a 18kHz lowpass should be generally very hard to detect. One should propably worry more about other oddities of the codec.

Here's another one, comparing uncompressed, Atrac3+ at 128kbps and Atrac3 LP2.

Atrac3+ at 128 kbps may have a lower lowpass frequency than LP2, but what you cannot see here is the awful artifacting of LP2.

[ozpeter]

Something worth trying - invert the encoded file against the original file in Audition's multitrack view - if they are identical, then you'll hear silence of course, but obviously lossy compression will never achieve that by definition. What you then hear is the character of the difference between the files, including artifacts. Perhaps look at the spectral view of that difference in respect of different bitrates. Personally I have never found it very hard to line up the two recordings exactly in time, which is of course essential.

Spectral view can also sometimes directly show what's going on with lossy compression - I recall seeing the spectral view of an mp3 file of parrots calling, and at some points you could see actual holes in the audio, where the compression had effectively dumped some frequencies altogether in an attempt to reproduce others it ranked as more important.

[Ishiyoshi]

@ greenmachine: The new test you have conducted - comparing ATRAC3plus 128kbps and ATRAC3 LP2 - really does gives one a different perspective on the new additional ATRAC3plus bit rates.

[Breepee2]

I hope everyone realises frequency analysis says absolutely nothing about percieved audio quality. The best thing is to do some ABX-ing yourself, or obtain ABX-results from a lot of other people and try to determine whether theres a trend or not.

[greenmachine]

We know - use flac only and you'll have no concerns about sound quality

[Breepee2]

Absolutely true of course

But when dealing with different kinds of storage (lossy Atrac3plus, FLAC, DVD audio) slapping a few lines over each other is going to tell you zero about percieved sound quality. And that's what count when you're judging a lossy codec.

[greenmachine]

Something worth trying - invert the encoded file against the original file in Audition's multitrack view - if they are identical, then you'll hear silence of course, but obviously lossy compression will never achieve that by definition.  What you then hear is the character of the difference between the files, including artifacts.  Perhaps look at the spectral view of that difference in respect of different bitrates.  Personally I have never found it very hard to line up the two recordings exactly in time, which is of course essential.

←

I did this kind of substraction in a different listening test, still doesn't tell much about perceived quality though:

http://forums.minidisc.org/index.php?showt...indpost&p=59476

[Alexx]

put it in simple terms for us

what does this mean *bounce*

Is ATRAC+ better than MP4

[raintheory]

I think it means, that whether or not Atrac3+ is better than Mp4 is entirely up to you. What do you think sounds better?

[Breepee2]

put it in simple terms for us

what does this mean *bounce*

Absolutely nothing.

Is ATRAC+ better than MP4

Since MP4 is a container format used primairily for the audio codec 'AAC', I presume you refer to the latter.

It's a thing you can best check yourself, altough the concensus seems to be (not necessarily on this forum) that Atrac3+ is one of the least interesting codec's today.

[greenmachine]

Considering its low power requirements it ain't too bad.

[dex Otaku]

I always kind of titter when I see tests like this. Aggregate spectrum analysis results when using music aren't really good for much. You would be much better off using sweeptones and complex noise to find where the filters are. In any case, except under the absolute worst circumstances [like a bug in the encoder or decoder], this kind of test actually tells you nothing whatsoever of the actual quality of the codec.

[Alexx]

how awful

[dewdude]

you also have to remember various rounding off functions are done in the frequency analysys in order to make the window more accurate.

Personally, i've never fully trusted frequency analysys overall...yes, 48kbps has a sharp 14khz cutoff...and it's ALMOST detectable...but...let's be honest...how many people have the headphones and the ears to tell?

I've tested various aac codecs at 48kbps and i've even taken ogg down that low in an attempt to find the best sounding format at the lowest bitrate...for a while my winner was 64kbps WMA....but...48kbps A3P does an overall nice job of shaping and masking the noise without doing a whole lot of damage to the stereo image.

there's sometimes a tad "something's a bit off" and the occasional moments where flanging is noticable...but...i think it's one of the best sounding low bitrate codecs for when you're trying to cram alot of music in a little space.

a 1GB Hi-MD disc at 48kbps supposedly holds 45 hours of music...i could put most of the good stuff from my music collection on a mere 3 or 4 discs......

tempting.

[syko]

There's a lot of things frequency response graphs don't tell, such as stereo replication and accuracy of sound to the original. Differences in the frequency response can easily be remedied through an equiliser, but nuances in accuracy, such as the dreaded "swishing" noise low bitrate files produces can't be eradicated (is it only me in this world that hears it??).

Stereo replication of low bitrate audio becomes a major challenge, but surprisingly enough ATRAC3plus does a decent job at 64k - it is noticibly better than standard ATRAC. This comes from a person that can hear the difference between 128 and 160k mp3 audio . As a rule of thumb, I generally try to avoid anything below 160k, but 256k is such a massive jump it's overkill

[exe]

This confuse me... When I wrote about HiMD in my best buy guide, did I write following, based on the information I found...

"HiMD LP mode give recordings of true MD quality (according to Sony) or slightly better than MP3 128kbps (according to deep listening tests) and you can then store amazing 34 hours... Atrac3plus has significantly lower sideband noise and reproduce the original frequency response, at every bit rate... So if the music is 15-22khz, will the copy be the same and not as 15-17khz with MP3 at 128kbps, or lousy 15-10khz with MP3 at 64kbps..."

But here was the higher frequency response, cut of at 14khz in HiMD LP mode and that is dramatically worse than MP3 at 128kbps, even if the frequency response may be of rather minor importance for the sound quality!!

So was this mesurement wrong OR was my old info wrong??

Edited December 20, 2006 by vantechmag

[exe]

Did just "revisit" the test results from Intertek Testing Services (UK) that clearly show a frequency response up to 22khz for both 64kbps and 48kbps Atrac3plus...

You find the PDF to the test here!

So had the test of this thread wrong settings OR did they change Atrac3plus dramatically to the worse, before release????

[cochra1]

you also have to remember various rounding off functions are done in the frequency analysys in order to make the window more accurate.

Personally, i've never fully trusted frequency analysys overall...yes, 48kbps has a sharp 14khz cutoff...and it's ALMOST detectable...but...let's be honest...how many people have the headphones and the ears to tell?

I've tested various aac codecs at 48kbps and i've even taken ogg down that low in an attempt to find the best sounding format at the lowest bitrate...for a while my winner was 64kbps WMA....but...48kbps A3P does an overall nice job of shaping and masking the noise without doing a whole lot of damage to the stereo image.

there's sometimes a tad "something's a bit off" and the occasional moments where flanging is noticable...but...i think it's one of the best sounding low bitrate codecs for when you're trying to cram alot of music in a little space.

a 1GB Hi-MD disc at 48kbps supposedly holds 45 hours of music...i could put most of the good stuff from my music collection on a mere 3 or 4 discs......

tempting.

Actually, most ears can detect a cutoff at 14khz. This is experienced as a lack of transparancy and lack of realism. That area of frequency between 10k and 20k is quite important when aiming for a relaxing, effortless sound.

However, I would say the bottom end is the more important of the two extremes to preserve - its the bottom end which is the first thing to be persevably lost when listening in most noisy environments (eg on a train) as man-made/machine/engine noises tend to generate a lot of low end frequencies which drown out your music's bottom end.