ATRAC3plus Developers' Interview

[Christopher]

Click here for bios on each of the four major developers.

— What is the history of ATRAC development?

Suzuki: ATRAC made its debut in 1992 with the launch of the MiniDisc. In 1999, as a new feature of the MiniDisc called MDLP, we introduced ATRAC3, which delivered an improved compression ratio. And just as electronic music distribution started becoming popular, we introduced ATRAC3plus in 2002.

— Is the technology behind these codecs very different?

Suzuki: The basic technology is much the same. All codecs take as their input a PCM sound source. They perform signal processing on a fixed number of samples which are converted into frequency data and then compressed (encoded). At the other end, these steps are performed in reverse, so the data is decoded. After restoring the frequency data, signal processing reproduces the PCM sound source. With each codec, there's a difference in the number of samples used for frequency conversion: the higher the compression ratio, the larger the number of samples.

— In developing the codec, what was the most important thing for you?

Suzuki: The balance between audio quality and hardware resources.

Inoue: The ATRAC algorithm is implemented in hardware, an LSI, so the goal of our development is a tangible product. To achieve this, it's necessary for the LSI team and the algorithm team to coordinate their development work. That's one of the major differences between ATRAC and other codecs – not many other codecs have been developed in tandem with hardware.

Tsuji: It's a huge burden on the LSI if you try to make it process each and every single task that comes along. So it's important for the algorithm and LSI development teams to find the point at which there is a practical trade-off. For example, when developing ATRAC3plus, the algorithm that I was in charge of threatened to be a severe headache for the LSI team, but after several meetings we worked it out.

Inoue: Yes, our premise that hardware should efficiently process ATRAC encoding is especially important for us. In many cases, the encoding for other codecs is based only on software processing on a PC. But an essential feature of the MiniDisc, for example, is recording from the hardware unit. That's why, when we developed the specifications for ATRAC, we had to make sure the LSI could process the encoding at high speeds and with minimal power consumption.

— What was behind the development of ATRAC3plus?

Inoue: We developed ATRAC3plus at a time when it seemed we had to answer a variety of demands – including both low bit rates and high bit rates.

Toyama: The ATRAC3 specification was originally developed to achieve a sound quality equivalent to the MiniDisc at 132kbps, so there was a limit to what kind of quality we could expect at a lower bit rate. But we applied technologies to solve the various obstacles we had faced in the past and focussed on crafting a new sound that would satisfy changing demands. ATRAC3plus is the result. In subjective evaluation tests conducted by third parties in 2003, it was confirmed that ATRAC3plus at 64kbps offers sound quality equivalent to MP3 at 128kbps.

Inoue: The demand for low bit rates started to coincide with the capacity increase of hard disk drives and other storage media, which meant, for some people, that a high compression ratio was not the most important criterion, as long as they could record with high quality.

Suzuki: Yes. Users who put more emphasis on sound quality than on the number of tracks.

Inoue: The answer for them is ATRAC3plus at 256kbps. CDs are 1,411kbps, so we are talking about 1/5 the size, but audio quality is almost as good as the original CD.

Toyama: In recent subjective evaluation tests, we proved that it's almost impossible to tell the difference in sound quality between ATRAC3plus at 256kbps and CDs at 1,411kbps.

The process of encoding and

decoding is repeated ad infinitum.

— You were talking about crafting the sound, but how do you go about optimizing or tuning?

Suzuki: Sound is subjective. Something about a particular sound may really grab your attention, yet someone else won't even notice it. When that happens, you get friends of like mind together and make that someone feel bad! (laughter)

Toyama: For example, people who play instruments tend to pay attention to the sound of each instrument, but your average person enjoys music as an overall experience. Everyone listens to music differently, so it's dangerous for just one person to do the tuning. I've played a variety of instruments, so when I listen to music I pay attention to how they sound, and the quality of vocals as well.

Suzuki: I'm the opposite. I listen to the whole thing. I belong to the generation that grew up with Walkman – music accompanies my activities. I don't notice the parts, but take in the whole song. As you can see, even within our team, we listen to music differently. But we don't give priority to one type of listening. It's important to find the best balance – that's our approach. So whenever I make progress in crafting sound, I get Tsuji and Inoue to join me in the listening room for several hours to get their opinions.

Listening room: here the

ATRAC sound is carefully crafted.

— What exactly is involved in this process of tuning?

Suzuki: We repeatedly encode and decode. In other words, we process the signal, compress it, reconstitute it and then listen to it again. Usually we use headphones, but when doing an overall check we pipe it through speakers in the listening room.

— Is there a difference between listening on headphones and listening through speakers?

Suzuki: Oh, completely different. To check details of the sound, headphones are the thing. But speakers are essential when you want to check the ambiance, harmonious sound, balance, and so on. You may think you've achieved a really beautiful sound, only to find that in an open listening space, that's not the case.

Tsuji: And because our constant tweaking is affected by our surroundings, we can't work in an open-plan layout to do our tuning. Our workspace isn't very large, but it's divided up with high partitions.

Suzuki: When tuning, we use a database of sound sources created especially for evaluation. For example, Mariah Carey is extremely tricky because the instruments and the vocals range over a very wide spectrum. And we can't make an encoder by just targeting a particular band. I've been listening to Mariah for 8 years – tens of thousands of times. Now I don't need to listen to the original sound source: just listening to the decoded sound I can tell whether it's good or not. I think her songs are indelibly impressed on my memory.

Testing equipment: provides

essential support for the

codec developers.

— And I understand that the results of this tuning are then evaluated by Sony Music (SME)?

Suzuki: Our work with SME started when they wanted to know whether or not they could use our codec for distributing music electronically. We asked them to assess our encoder and found them to be incredibly tough critics! (laughter)

Toyama: SME distributes music electronically and whatever codec they use, they're responsible for delivering the end result to paying customers. Obviously it's no good if the sound quality doesn't satisfy SME. That's why we've worked hard to tune the codec to win SME's approval.

— The codec used in SME's online music distribution is ATRAC3?

Suzuki: That's right. The sound we created around 2000 is still being distributed today. But there were times when the development team and SME didn't agree. Developers tend to emphasize details, but SME values an overall balanced sound. We use sound sources they give us for tuning, after which recording staff at the SME studio test the codec rigorously for things like sound localization and coherence.

Toyama: The sound sources from SME are their own special mix. They know it front to back, exactly what sounds are where. So if something is even slightly off, they know immediately.

Suzuki: Because you're compressing sound, it can't be helped if the result doesn't sound the same as the original. But it's not acceptable if the compressed sound changes the impression of the original, so balance is especially important.

Toyama: That's right. We look closely at how the compressed result sounds. Even with compression, people want the music to sound good, if not great.

Suzuki: To be precise, while admitting ATRAC3 compressed sound is not the same as the original, we've arrived at the point where the way in which the sound has changed, is good – good enough for the ears of the recording staff at SME.

— What about ATRAC3plus?

Tsuji: SME has yet to evaluate ATRAC3plus. Originally we targeted our development for compact media such as flash memory. So we focused on high compression ratios and worked hard to craft good sound quality, even for bit rates as low as 48kbps and 64kbps.

Suzuki: If the bit rate is high enough, there's enough data to cover any shortcoming even if the encoder doesn't work exactly as hoped. But the lower the bit rate, the more likely it will be affected should the encoder performance be just a little wanting – it would appear as noise or otherwise sound poor. That's why we spent a lot of time enhancing the 48kbps and 64kbps encoders. We're confident that they sound better than any competing codec. In the future, with the cooperation of SME, we want to make audio quality even better.

— Recently, the rising popularity of personal audio players equipped with hard disks has increased the interest in codecs. As codec developers what would you like to see in ATRAC3plus-compliant products?

Inoue: For our part, we will continue to work hard on codec development, and we'd like product developers to continue to focus on crafting the quality of the sound at the point where it is converted from digital to analogue.

Suzuki: Of course, product developers are concerned with the whole player, but I think it's an advantage for audio manufacturers like us if we can optimize each of the components that make up a player. As a user myself, that's what I expect.

— As ATRAC developers, do you have a message for the readers?

Tsuji: Some people seem to have the bias that "compressed sound is bad", which I think is a bit of a shame. Maybe it just comes from familiarity with linear PCM compact discs – their very presence helps create this impression of inferiority. I really hope that people put aside any prejudice they may have and listen to compressed music with an open mind.

Toyama: I once worked in an SME mastering studio for a month as part of my production training, and was surprised to find the vast number of sounds in a CD recording. If you compare compressed music to that, of course it isn't the same, but it does create new ways to enjoy music, such as online music, and lets you store a lot more songs on recording media. Whether we're talking about a natural shift in sound quality, or being able to enjoy music for hours without getting tired, what we do, we do for music – after all, we love music. And we'll keep on developing to expand the possibilities of sound.

Suzuki: When we created the ATRAC3plus algorithm we put a lot of emphasis on high quality sound. So I would ask the readers to experience it first-hand, and at different bit rates.

Inoue: Yes, our ultimate goal as developers is to have listeners appreciate just how good the music sounds without noticing the bit rates. We will continue to work hard to craft the kind of sounds that allow people to enjoy music naturally and with the greatest pleasure.

[dex Otaku]

This makes me wonder if these guys have listened closely to the differences between the Sony Professional atrac/3/plus codecs and the one that are part of SonicStage.

The LSI version will end up being the reference, sure, but these guys must do a hell of a lot of work in software only.

It's nice to hear from the other side of the fence for a change.

[Christopher]

I have a better outlook on ATRAC3/plus after reading this.

And yes I agree - if Sony was at least more open with development it would attract interest. The reason why codecs like flac, ogg, aac and so forth are so popular is because the community revolving around it had a large influence on it's direction - something that ATRAC3/plus doesn't have. Sony's close-minded nature really astounds me compared to the days of old, but perhaps later this decade they'll have either imploded or blossomed / reborn into something grand.

Maybe we should take bets.

[jadeclaw]

A problem with a low number of test-listeners is that the personal listening preferences can go into the codec.

With OGG, it was different. Due to its open development process, it had a lot of test listeners, averaging out any personal preferences, that may exist.

That's why OGG is that good, especially at low bitrates.

Sony should start opening up the development process, eventually even placing the ATRAC3/3plus codec under the GPL, thus allowing users to make improvements going back into SonicStage and other Sony products.

Trust me Sony, the many ears approach works...

[aeriyn]

Well that was certainly interesting. Too bad we can't trust much of what they say since they're biased.

[nitpick]ATRAC3 132kbps is nowhere near equal to Type-R ATRAC. Not even close. I'd say LP2 is equal to ATRAC v1.0. That'd be a fair comparison.[/nitpick]

[Mr Sim]

Anyone Else bnotice that in one of the photos the programer is using a mac with OSX yet the common consumer products from sony are not mac compatable.

[Daijoubu]

Power consumption is important too, who use uncomp wav with thier ipod?

Say good bye to your battery

[hobgoblin]

Anyone Else bnotice that in one of the photos the programer is using a mac with OSX yet the common consumer products from sony are not mac compatable.

←

hmm, im guessing its a sound techie checking the quality of the output sound from the hardware rather then a programmer working on the atrac code. allso the equipment stack to the right of the person is the same as shown in the image under the SME heading, again pointing towards a sound techie (some SME most likely) doing quality testing of the output...

Edited March 1, 2005 by hobgoblin

[Seikeden]

It's interesting that they care so much about hardware performance/power consumption issues and the necessity for an effective hardware solution.. yet it's like they hired a team of %$#* monkeys to do the pc side of things.. (which to be quite honest, is EVERYTHING to hi-md.)

sorry im in a rant mood, I just found out sonicstage alters mp3 files it indexes, so they fail hash checks.. AND one of my older recordings saved in lovely oma format is now apparently incompatible with the new sonicstage GRRRR

[greenmachine]

Why is there no medium bitrate (about 128 kbit/s) anymore with ATRAC3plus?

[zerodB]

Well, as Sony sees it, ATRAC3plus is merley an "extension" of the ATRAC3 format:

ATRAC3plus is the latest development in codec technology; while keeping compatibility with ATRAC3, it compresses audio data to 1/20 the original CD source and maintains extremely high quality sound.

Edited May 20, 2005 by zerodB

[greenmachine]

They call 64 kbit/s Atrac3+ Hi-LP extremely high quality sound? LOL

It's nowhere close to an 128 kbit/s mp3, even 132 kbit/s Atrac3 LP seemed to be (slightly?) worse as i compared it directly.

edit: I've set up a comparison, see this thread:

http://forums.minidisc.org/index.php?showtopic=10541

Edited June 7, 2005 by greenmachine

[ROMBUSTERS]

a

Edited January 9, 2008 by ROMBUSTERS

[FlameGrilled]

Well, Rombusters, i agree totally with the sentiments about the sources and codecs and compression in general.

I added a note in a different topic that the audio source really defines what's needed (in a perfect non-world of transparency), and it's all a balancing act on a good day getting results of encoding.. let alone getting a codec to work efficiently and up to some kinda standard of output.

There is stuff i listen to, on audio CD or vinyl or cassette (1st gen studio grade stuff dumped onto consumer media for my purposes) that literally i would (if i was audiophile or quality obsessive) would say don't encode very well in any lossy codec (even 500K Ogg output can sound a little tainted for my minor picky moments) - but when i keep in mind that DAP playback (in lossy circles) is really an arena that's more comparable to listening to CD transferred to HQ tape type transfer (like it's analog analogy, but lacking the noise clearly and without the CTFR limits), it kinda tells me that (out of knowledge and experience) that transparency is a kinda lost cause.

It's simply a case of use what works best for each audio instance, whatever combo gives you usable results (unless you are doing glorified walkman mode, in which case it's SP/LP all the way), and don't be afraid to mix and match to get the desired combination of content stored at a reasonable (read useable) quality.

Interesting interview mind..., tells little.. but is usefully that extra step more useful than the average Sony blurb

----------------------

Former MD SP user (back in my radio days).

Defected to mp3 and WMA...

Returned to MD recently, after being asked to master demo's and short-quantity MD and ATRAC CD releases, so i'm using top-models from each side of the personal portable ATRAC world, as personal devices and to act as portable demo machines.

MZ-NH1

D-NE1

D-NE20

Other stuff :-

iRiver H300

iOPS MFP-350

iRiver iMP-250

iRiver iFP-390T

Archos AV480

The most unique, custom-build DAP-64 (x64 based transportable 24 track integrated editor/recorder/processor/mixer/player - supports all non-ATRAC formats, hopefully may support ATRAC one day).

And a crap load of ancient pro analogue and digital stuff, incluiding portable Betamax recorder with digital adaptor/mod/demod.

[AMPlitude]

very very interesting interview... atrac sure has come quite far since introduction...

[Tunster]

I know this post has been laying with no replies since the end of 2005, but I want to put my $0.02 in.

After reading that, ATRAC has had a very interest side of closed-sourced development. In their eyes and ears, the 64kbps and 48kbps, its been a great innovation. I will agree with them. You encode a Stereo 48kbps/64kbps MP3 and you get trash out the other end. They've just fined tuned it to be "listenable", which it is. But to the recordist and audiophile, it isn't what they make it out to be.

If Sony ever put in the white flag with MD tech, I wouldn't mind them releasing some SDKs of the ATRAC codec and firmware of ALL of their MD units. Would be nice to see some open source coders get a can opener on it and see how much further you can push ATRAC....I reckon some coders definately could push it to AAC standards. That would really turn on the continuity of MD and maybe relive it? I'm being hopeful, but it would be nice to see one day.

Edited June 10, 2006 by Tunster

[atrain]

If Sony ever put in the white flag with MD tech, I wouldn't mind them releasing some SDKs of the ATRAC codec and firmware of ALL of their MD units. Would be nice to see some open source coders get a can opener on it and see how much further you can push ATRAC....I reckon some coders definately could push it to AAC standards. That would really turn on the continuity of MD and maybe relive it? I'm being hopeful, but it would be nice to see one day.

odd that you should say that, there's been some talk although with little or no development on that score

[Tunster]

odd that you should say that, there's been some talk although with little or no development on that score

From that perspective, it would be logical for Sony to do this. Especially if they want to keep ATRAC compatibility in their products (e.g. PSP + their HD range of players).

In terms of the MD hardware/firmware, it would be great to start adding even more features to MD decks/portable recorders to relive their life expectancy. I just don't want programmers doing ugly reverse engineering to MD hardware, and Sony probably eouldn't want this either.

Atrain: have you any source of this talk of Sony doing this? Would be nice to read up.

[cyberhill]

Thanks for the interview, it was very interesting to read.

[girish]

very very interesting interview... atrac sure has come quite far since introduction...

Guys it is not bad....

I checked it out... on D-NE715

Music quality is fine ...

Good when u r going on trip...

Can get more songs on 750Mb disc...

[pata2001]

Sony Music execs look at Atrac for the DRM, highly unlikely for any "sound quality." Also consider connect store still uses the outdated LP2. With more and more support for AAC and HE-AAC, I can see that is the direction Sony is going (DRM compliant, highly supported in many devices like cellphones, and good quality at low bitrate provided by HE-AAC).

[cochra1]

Very interesting interview. Totemo omoshiroi intabyuu!

Edited February 12, 2007 by KanakoAndTheNumbSkulls