PCM vs. Atrac 3 256k (quality comparison)

[jadeclaw]

+0.145 better. Yeah, that's a huge margin. (sarcasm)

In reality, MP3 @ 128 k is a lot better. One warning tho, don't use the Xing-Encoder for MP3-encoding,

it is as fast as diarrhea and the result sounds like diarrhea -

Use LAME or Fraunhofer instead.

PDOG, but despite my acidic criticism of Hi-LP, Hi-SP sounds very good and the difference to the uncompressed original is so marginal, it can be considered transparent. So, no worries here.

... and then you advertise that "maximum recording time" right beside your claims of high fidelity.

When comparing, that was my first thought...

[jadeclaw]

Try iceeedtea's build for me. I've never had an issue with it.

Update!

SonicStage killed my WinXP installation in a very peculiar way.

During boot, the screen goes blank, the Sync for the Monitor is switched off while the harddisk light is lit permanently.

The lockup was so bad, that even the reset button doesn't work. I've had to cycle mains supply to unfreeze the machine.

Sometimes 2 or 3 times.

Last time I've seen this was on a Commodore C64 with a virus...

Now, after reinstalling everything execept SS2, the problem is solved.

[Thorgal]

Hello,

This is my first posting here.

I make use of Minidisc since 1997, have an deck in the living room, car equipment and a Net-MD. I like Minidisc mainly because of its size, but I have always had questions about the sound quality.

I am very intererested in the several codecs and their soundquality. I think it is interesting to see how people or orginisations succeed in making files smaller and/or improve the quality.

Two things surpised me lately. The first thing I mentioned already and that’s the quality of Atrac 3. Second is the quality of MP3. More often I read that the mp3-codec is old and that real high bitrates are needed to get an acceptable sound.

My experience is a different one. I think MP3 sounds okay when its encoded with Lame at 160 or 192 kbits. My equipment at home is very good, I own a Harman Kardon amplifier. I enjoy listening to it, its relaxed with hardly any artifacts. Ogg gives me already very good results at 128 kbits.

I wish I could say the same of Atrac 3, but it just can’t. Listening with my headphones gives me the best result (how strange). But listening in my car is sometimes really bad. I have minidisc mixed with several codecs and just be listening I can tell which codec is used, SP, LP at 132 or LP at 66. LP at 66 is really bad, but differs from the music. The same goes for 132, but that is less worse. Some people say it sounds good. When I read that I can only think: are you deaf or so? No offense, but it’s really what I think. SP sounds fine, but no wonder at 292 or so.

Interested in codecs as I am I was very curious about Atrac 3 plus and I had great expectations. Depending on the results I would love to buy a Hi-MD.

Now I have done some testing with Sonicstage 2 and I am really disappointed. I hear almost no difference with Atrac 3.

Try for yourself and encode the latest cd of Anastacia to atrac 3 of atrac 3 plus at 64 or 132. It just sounds horrible. Mp3 als gives no real satisfying results but is much better then Atrac.

In minidisc forums all over the world you can read people telling about the superior quality of atrac compared to mp3. I always found myself alone in thinking that Mp3 just sounds a lot better.

Now for the first time results about atrac 3 compared to other codecs have been published by Roberto Amorim. His test are anonymous and blind and world –wide taken. He is seen as objective and is seen as an authorization.

You can read the testresults here

http://www.rjamorim.com/test/multiformat12...28/results.html

The test show clearly that Ogg is the best and Atrac is definitely the worst codec around.

Finally is get some backup here!

My conclusion: I stick to Minidisc but will not replace the stuff when it is broken, I will not buy Hi-MD. Sony has to do some work first on their codec. But I wonder if Atrac will ever reach the quality of Ogg or MP3. Explination is simple: on mp3 or Ogg are working lots and lots of people, Atrac is only used by a very small minority. Sony has to increase sharehouldersvalue and are the only ones to invest in the codec. They just will not be able to compete with the others.

This is what I expect and I think it is a pity. I already said I love the minidisc because of its size, but also flexibility.

[polario]

I find it interesting that Roberto Amorim chooses to mix CBR with VBR encoders in his Multiformat at 128kbit/s Listening Test. Which means some of the encoded material contains more than 128 kbps of compressed audio in it's data stream. No wonder the variable bit rate encoded music was percieved as better quality.

The results clearly show as expected that music encoded with higher average bit rates correlates to better quality, with some exceptions due to compression algorithms used.

For example the song Waiting by Green Day received the following results in the listening test:

4.64 - MPC (153 kbps)

3.95 - Vorbis (144 kbps)

3.76 - Lame (148 kbps)

3.67 - iTunes (128 kbps)

3.65 - WMA Standard (131 kbps)

3.49 - ATRAC3 (~128 kbps *)

From the list above one can only directly compare iTunes with ATRAC3, of which iTunes is percieved as producing higher quality results on this particular piece of music.

I dare say this test seems unfair and to some extent misleading!

[jadeclaw]

Creating equal conditions was the first thing I did in my tests.

With Nero's MP4-AAC it was simple, select CBR, then the closest to what Sonicstage offers == comparable.

With OGG Vorbis, it was more complicated, nailing it led to error messages, but with a bit of playing around with the different settings, I came quite close to the desired bitrates - much closer than Amorim in his. And I usually ended up slightly lower than the SS2 ATRAC3 offerings. But despite that, OGG and AAC performed better.

Nevertheless, Amorim devaluated his test by not paying enough attention to the test conditions.

[Thorgal]

hi,

I've done a scientific education myself, and there's one thing I can tell you: the prefect test just doesnt exist. One can only strive for the perfect test, but it just can't be achieved. It's an illusion, and that is one of the reasons why scientific research, afters decades, still goes on, and always will.

Roberto Amorim hasn't reached the status of authority for nothing. His test comply with the standards that are used in scientific environments. That is why his test have such an impact and are often linked to.

You can be sure that this report lies on deck of many exectutives at Sony and that some people at Sony have something to explain right now and have to take action. Specially considering the launch of Hi-MD.

It is not fair of polario to take just one song and set that as an example. It is the other way around. Some codecs just don't offer VBR, which may be considered as a minor point of that codec. In this case Atrac 3. Because is is not the bitrate that is important, it's the size of the file that is!

To make an honest comparisation between the codecs Roberto has solved this problem by taken 18 songs. That way he offers a solution for a weakness of some codecs, which don't offer VBR. Polario uses this weakness against the test! After all it is leading to an acceptable average bitrate for the 18 songs in total:

Averages are :

iTunes: 128

MPC 136

Ogg 135

Lame 134

WMA 128

Atrac3 132

Difference are to be considered small and acceptable.

Fact is that Atrac 3 is considered by far the worst codec around, and Ogg is the best. I am just as surprised as many others. Lots of professional sites regarding audio offer links to the test and show their surprisement. But on the other hand I am happy that there's nothing wrong with my ears! It just is a pity that few hardware support Ogg Vorbis!

[polario]

Thorgal, I'm happy you have shown such a verbose interest in audio compression technology and comparisons thereof. Although I do respect your opinions, I beg to differ on a few points.

The ATRAC encoder has evolved over the years, from a not-so-hi-fidelity start to the perfectly enjoyable SP mode quality of today. In a quest to increase recording time, Sony sound engineers came up with the LP2 mode (MDLP). For backward compatibility reasons, this new mode uses the same 212 byte sound group rate (padded with 20 bytes of dummy data) but twice the transform window size of SP.

Compatibility is the reason ATRAC3 LP2 mode is a constant bit rate (CBR) encoder. It is my understanding that to all intents, constructions, and purposes, it could easily be made into a variable bit rate (VBR) encoder. But this is not the point of my argument.

First of all, to compare CBR to VBR and from that decide upon the superiority of the compression algorithms is very odd indeed, only proving the tester's level of knowledge and comprehension of the technology: On one hand, a CBR encoder frequently needs to discard some data provided by the audio compression algorithms in order to meet a fixed data rate. On the other, what VBR encoders do is to make unused data rate available to transform windows needing more space in the data stream.

So in effect, what you get is apples shoved through slightly too small steel pipes (CBR) and oranges through rubber pipes (VBR). You don't get to compare intact apples to (undamaged) oranges. Which would be very interesting to do.

In an audio fidelity test where the smallest common denominator (in this case the data bit rate) is not equal, the results can not be an indication of the quality of the audio compression technology used in different encoders, but only the subjective opinions of the test group regarding music in various data rates.

As to acceptable variance in data rates. Comparing encodings of a track with data rates ranging from 128 kbps to 155 kbps is hardly acceptable.

Regarding the quality of audio encoded with differents compression algorithms, I am not disregarding the fact that there are more sophisticated encoders than others. I am merely pointing out that, in my firm opinion on the matter, the test is unfair and I can only disregard it as useless.

Clearly Mr Amorim has may other irons in the fire. He should not hesitate to put this test with his other irons.

- -

On the number of examples to use in discussing a topic: Using more examples would only prove to be redundant, as a link to the original test was supplied. Bearing this in mind, I cannot refrain from quoting George Latimer: "When I open an egg, I don't have to eat the whole egg to discover it is bad."

As a frequent reader of forums, I tend to value concentrated and clear postings. Although, exceptions make the rule and I do hope the forum readers will forgive this slightly off topic rant. :grin:

- -

Regarding ATRAC3plus Hi-SP mode vs PCM: While doing a A to B comparison, with good listening equipment, there is a difference in ambience, but otherwise Hi-SP is more than good enough. In other words, I will be perfectly happy with Hi-SP for my afternoon jog around the park, but at home I will continue to enjoy my music uncompressed.

/Polario

[Thorgal]

Hi,

Regarding one your remarks I shall keep this posting both short and clear. We wíll not agree about the quality of the test. Like I said many professional websites about audio regard Roberto Amorim as an authorization. If you think this test sucks, you are in a very small minority but you are entitled to.

Quality of testing was not my subject, quality of atrac3 (plus) at 64 or 132 is! Test or no test, atrac 3 at 132 or 64 just sucks. To make an equal comparisation: Ogg in CBR at 64 or 132 sounds a lot better. Not only Ogg, but several codecs I tested myself over the years. And quality is all that counts to me. Bad test or no bad test.

[polario]

Being regarded as an authority of any field these days apparently does not seem to require much. A Scientific study wanting to be taken seriously, should address all aspects of structural weakness (in this case); lack of smallest common denominator, statistically sane number of input data and in its findings produce an error margin accompanied with appropriate analysis of all aspects.

Mr Amorim's test consisted of several different music types, which is good, but fails in getting more than 27 inputs for any of the tracks tested. The results of a statistical study with anything less than 100 participants can not be viewed as representative of the general public. The statistical base is simply too small. Why do you think opinion polls use 1000, preferably 2000 participants? They would cost far less to ask only 15 people their opinion, but then again the poll would not have any value.

Most importantly such tests should always be performed in a controlled environment, which it was not.

As to the quality of ATRAC: Recording modes other than SP, can only be regarded as what they are, ie, marketing gimmicks by Sony to increase (double/quadruple) recording times on the minidisc. They work reasonably well.

Digital recordings (SP) I have made with my Sony MDS-JA333ES home deck (ATRAC with Type-R DSP) sound extremely good even compared to the original CD. SP mode is what ATRAC is all about. It was designed for this bit rate and very few encoders can compete with it. If you are out of space, buy another minidisc, do not use LP2 or LP4 and complain about the quality. I have yet to find a unit with another codec using portable media that will come anywhere near quality wise.

- -

My only concern about the new ATRAC3plus at Hi-SP is if it will sound as good as with my current equipment. It remains to be heard.

With the new Hi-MD units one can get 7h 55m of Hi-SP or 1h 34m of PCM on a single disc. Having hours worth of good quality music can be practical because one can have an entire multi disc album on a single minidisc. But what is the point of having days of music that you only listen to in the background. And why demand hi fidelity elevator music?

Music should be enjoyed, not faded into the background.

/Polario

[anont]

The results of a statistical study with anything less than 100 participants can not be viewed as representative of the general public. The statistical base is simply too small. Why do you think opinion polls use 1000, preferably 2000 participants? They would cost far less to ask only 15 people their opinion, but then again the poll would not have any value.

Many medical studies have FAR less than 1000 participants, and can still be considered statistically significant. The statistical technique is to show a margin of error based on how small the small sampling size was - and that's what Mr. Amorim did.

Digital recordings (SP) I have made with my Sony MDS-JA333ES home deck (ATRAC with Type-R DSP) sound extremely good even compared to the original CD. SP mode is what ATRAC is all about. It was designed for

I agree with this - who uses 128kbps? Now that hard drives are cheap, this doesn't seem like the most useful study. SP does sound very good, if not quite CD quality. On the other hand, so do the other compression formats at a similar bps.

[dlo]

I just wanted to say that I feel NetMD LP4 does not suck, I have sold my NetMD already and am desperately waiting for HiMD to enter Europe and I will probably be recording @ atrc3+48kbps AND it will sound just fine, I'm sure.

And another thing: I am also sure that there IS a PERFECT audio test - these are all waves mr. scientist and they can be represented on a screen and different transformations can be fully analysed.

So I keep wondering: who the hell has the perfect test results?

Sony obviously hasn't got 'em, because they work with a group of human listeners who think at the end of the day about what confidence intervals they are gonna give their results. Also any computer result based on PC atrac3 codecs is also not accurate, it should be from the HiMD itself.

So which fantastic Oriental person can post the ultimate test?

[jadeclaw]

And another thing: I am also sure that there IS a PERFECT audio test - these are all waves mr. scientist and they can be represented on a screen and different transformations can be fully analysed.

And exactly that's, what doesn't work with lossy encoders.

Remember, 80% of the data is thrown away during encoding(Hi-SP). Much more on lower bitrates. And that explains, why comparing waveforms doesn't work.

Since the human ear does the same, the key here is to match the encoder with the characteristics of the ear.

So, only listening tests work here.

So I keep wondering: who the hell has the perfect test results?

Noone has them. And no one wants them. We don't need perfect test results, we need the perfect test conditions to get objective results.

And that means: Double blind test. Like it is done with pharmaceutical substances. And that quickly becomes a problem, as that needs time, a lot of time.

And has to be repeated, whenever the encoder software is changed. So there is a point, when it becomes too expensive.

The second best way is what I have done - matching bitrates and using the same soundcard and external D/A converter for all of them.

Also any computer result based on PC atrac3 codecs is also not accurate, it should be from the HiMD itself.

Well, Yes and no. First, when testing software encoders, it is important to use a non-resampling soundcard, as with resampling cards the soundquality depends on operating system and drivers.

Second, with SonicStage, what is encoded on the PC ends up on the MD. No reencoding is done on the recorder.

Which brings us to the final question: How small are the differences between SonicStage and the final hardware?

And in that case, we have to wait for the recorders with analog/digital inputs, so that we can check the whole chain.

My tests aren't final as well, the encoders in the next SS2-version could be different, and the encoders in the recorders change with each chip generation as well, so later units can sound better - we have seen that in the past...

[Thorgal]

If I read people saying that LP4 does not suck and that they will probably record in 48kbps, and that it will sound just fine, I can only repeat that one must be almost deaf. Are those the same people who say that Atrac is superior to Mp3? It would explain a lot.

I totally agree with what Anont writes.

As told i am interested in codecs. To produce CD-sound at 256 is no hard job for any codec. Of course Atrac sounds fine at 256, but so does MP3 at that bitrate (Lame > 3.00). But at 128-160-192 it IS getting interested: that's where differences can be heard. Sound quality at lower bitrates gives information about that particular codec. For me the test is a confirmation of something I have been hearing with my ears for years now: end of the story that Atrac is superior the Mp3.

[Sbetsho]

about the tests.. when I listen songs in LP4 mode with the earphones that came with the minidisc, they sound okay, but when I tried other headphones, they sound horrible.. perhaps this was a factor in those tests.

also, when I hook my minidisc on stereos and play LP4 songs through big speakers, it sounds just fine, altough I'm not able to listen very loudly because of my neighbors.

I then might be deaf, but I don't notice much difference between SP and LP4.. LP4 sounds flatter and not so dynamic, but it doesn't suck. but then, I'm not used to hi-tech devices and hi-fi sounds, like most of you apparently do, so I understand that you think that even LP2 sucks (for me it sounds every bit as good as SP)

I heard that there are people that thinks even Linear PCM quality sucks.. now that's insane.

But then I guess I'm that kind of type, that doesn't care so much about the sound quality, the songs are almost as enjoyable in LP4 (it certanly doesn't sound 4 times worse than SP) as it is in SP quality (but not much worse than LP4 is too bad even for me, and I hope 48kbps Atrac3plus isn't that bad)

[Damage]

Well, all stats aside...

If you "discount" couple of the samples (the Techno one seemed to trip up ATRAC big time), statistically speaking, the ATRAC does no worse or no better than other CBR codecs, the WMA and the AAC.

As for the VBR codecs, the setting was chosen, as I understand it, such that the overall bitrate from GBs of music would be near 128kbps (or 130ish there abouts). However, the samples that were chosen weren't really reflective of it, unfortunately. Perhaps some of the lower bitrate music were chosen as to pad the results? If there were indeed GBs of samples encoded as such that the final results over all of those were near 130, how hard would've been to pick some that were really reflective of those bitrates?

As I said it seems like the samples made it such that you're comparing Apples and Pears, not quite Apples and Oranges, but Apples and Pears.

By the By, Sbetsho, I consider 48kbps no better or no worse than LP4, and 64kbps A3plus to be better than LP4. So use your judgment on that. I happen to like HiLP (64kbps) and LP2 and HiSP sounds pretty damn nice.

By the by, most competent codecs will produce CD transparent files at or around 256kbps, unless you're talking about Blade. Blade won't achieve transparency, evar.

[highlandsun]

For example the song Waiting by Green Day received the following results in the listening test:

4.64 - MPC (153 kbps)

3.95 - Vorbis (144 kbps)

3.76 - Lame (148 kbps)

3.67 - iTunes (128 kbps)

3.65 - WMA Standard (131 kbps)

3.49 - ATRAC3 (~128 kbps *)

The above note is wrong. ATRAC3/LP2 has a true data rate of 132kbps, not 128kbps. The dummy padding only exists on physical MDs, and would make the disk-to-memory data rate 146kbps (same as mono-SP).

[Sbetsho]

By the By, Sbetsho, I consider 48kbps no better or no worse than LP4, and 64kbps A3plus to be better than LP4. So use your judgment on that. I happen to like HiLP (64kbps) and LP2 and HiSP sounds pretty damn nice.

Wow.. thanks.

[Leland]

I ripped a CD at ATRAC3plus 66 and the same one at ATRAC 3 133. I was surprised at how good the plus 66 sounded. Looking forward to trying it on Hi MD's with lots of music.

[Jimmix]

I really liked NetMD Atrac3@132. So if what SONY tells in their comparison chart (that Atrac3plus@64 will sound like Atrac3@132) is true that would be great.

Now by what a lot of you guys have been saying that's not quite true.

However I wonder if the HD dig. amp. present in the 900 will help Actrac3plus@64 be heard good enough???

For me good enough means no easily heard artifacts.

[jadeclaw]

I don't think so. A better amp will reveal more artefacts, so the difference will be more audible.

The only hope I still have, that the encoder in the recorders is better than SonicStage. But that means, realtime recording.

Until that is out, I suggest to stick with LP2@132k or better. In addition, you can use that in Hi-MD mode as well, which in itself nearly doubles the playing time per disc.

[Jimmix]

Ok.

Now how about the quality... I mean you guys that may have more experience...do you know if Atrac3plus@64 has audible artifacts just like Atrac3@66 has?

[mmp64]

To my ears, absolutely - ATRAC3Plus 64 (Hi-LP) has artifacts. I believe it is superior to the old LP4 - but still not good enough. I'm sticking w/ LP2.

[Christopher]

Good call, mmp64 - I think I'll be primarily employing Hi-SP on my discs, though. You'll get good times with LP2 and Hi-MD, nearly 5 hours with a formatted regular MD into Hi-MD mode disc, or 16 1/2 hours with a Hi-MD disc!

[brandon_ae]

damn, Im a huge Minidisc fan, but after reading this entire thread and coming to the conclusion that Ogg is the best and Atrac3plus is just "good enuff' I want to go out and buy an ipod mini for my entire VBR mp3 library. Then again I dont wanna fill up my 4gigs and have to delete some just to fit more on, even a 40gig would fill up eventually. And by the time it came to delete files, u might be interested in the song again cuz u might not hage heard it in forever if ur using a 40gig. I think I'll just suck it up and stick with the Atrac3 plus 256k just for the fact I can carry multiple 1gb hard drives.

[mmp64]

After doing a lot more testing, I've decided to go w/ Hi-SP for most of my stuff - or, at least the stuff that warrants it (newer recordings, remastered recordings, etc.) LP2 is very good - but Hi-SP has an intangible "feel" to it that makes it sound fuller - wider - whatever. (not surprising, given it is 256K).

So the way I look at it now - Hi-MD (at least using 80 min discs) lets you get SP quality at NetMD LP2 quantity, more or less.

[Andy]

One of the reasons I'm going to give ATRAC3 (and plus) some slack is that I believe that it is designed to use less battery power than other formats when decoding (and presumably encoding too). Ogg Vorbis is a file format which uses a fair bit more processing power than previous generation formats (like MP3).

On the other hand, the big eater of batteries is just the size on disc (look at battery life figures for PCM mode).

I got my brother to blind test me on CD and ATRAC (SP mode, as recorded by my MZ-R70), and while I could pick the difference consistently and easily, it was still very acceptable (most of the artefacts were just a lack of definition in the bass)

Ogg Vorbis is great (I use it for everything on my computer), and I would love it if Sony would use something like that for MD. Unfortunately, they have created a format which is probably inferior (though I haven't tested it), but will still be acceptable.