I always enjoy and appreciate Ethan's presentations and challenges.

One of the recent discussions on one of the pro recording forums was the audibility/necessity of dither when truncating from 24-bit to 16-bit.

Ethan's point is that dither makes no audibility on the majority of recordings recorded at normal volume. This would include essentially all pop, rock, etc.

Only one example of dither audibility was offered. This was a live jazz recording. During the quiet passages (-55dBFS or so) one can hear the difference between the dithered and non-dithered file.

I have also heard differences in dither in long reverb tails, especially natural reverb (not from a box or plug-in).

I think Ethan is right, dither rarely matters but there are isolated incidences when it does.

However, it is essentially free so I always add dither when reducing the bit-rate of my recordings (typically iZotope's MBIT+ 64-bit dithering algorithm) just in case somebody else can hear it.

Quote:
this is the video version of Ethan Winer's AES presentation.

Geared more towards the recording engineer, but some here may find it interesting. I know I did.

http://www.youtube.com/watch?v=BYTlN6wjcvQ

Hmm, my ghu, who was the ugly old guy who did the first talk after the intro?

I'm sort of annoyed that all the meat is removed, but so far, the editing hasn't changed the message, it's just removed all of the "why this happens".

Quote:
Hmm, my ghu, who was the ugly old guy who did the first talk after the intro?

I was wondering . . .

Good to hear tINY's voice also.

Quote:
During the quiet passages (-55dBFS or so) one can hear the difference between the dithered and non-dithered file.

But only if you raise the volume unnaturally high to hear it, yes?

Quote:
I have also heard differences in dither in long reverb tails, especially natural reverb (not from a box or plug-in).

But only if you raise the volume unnaturally high to hear it, yes?

Ethan Winer
RealTraps

Quote:
I'm sort of annoyed that all the meat is removed, but so far, the editing hasn't changed the message, it's just removed all of the "why this happens".

From my perspective it made sense to keep only "this is what happens" because the "why" got pretty technical and would have made the video two hours long instead of only one hour. I thought you were well represented, and I loved your tube versus transistor "clack clack" anecdote.

Ethan Winer
RealTraps

Quote:
But only if you raise the volume unnaturally high to hear it, yes?

Yes.

Only if I listen to that portion of the file, isolated, and at a much higher level than I would listen to the entire file.

But it is there - which I know that you too acknowledge - and thus I use dither. The odd noises then go away.

I think you are correct. A file truncated from 24-bit to 16-bit without the prior application of dither contains artifacts that we can measure, but at such low levels it is masked other than in isolated circumstances.

On the other hand, I think it is cool that I can hear it under these isolated circumstances and I am delighted to have the tools to make it go away.

Plus, if someone else listens to my recordings that can hear a non-dithered bit-rate reduced file I don't want him or her reacting with disgust.

Hence I dither.

Proudly.

I don't disagree. Dither is free and can't hurt. But as I explained in my video, claims that dither affects fullness and imaging etc are wrong.

Ethan Winer
RealTraps

Quote:
But as I explained in my video, claims that dither affects fullness and imaging etc are wrong.

I don't get this either. I also find troubling (amusing?) that some say they hear the difference in the lows; while others claim, the highs.

This simply is not what dither does.

As you stated, it doesn't hurt but it is hardly magic that will polish your recordings if properly applied.

As an aside, I love the posts asking what dither to use. I am always tempted to reply that if you can't hear the differences between various forms of dither you need not worry about it.

Some day I will play with 8-bit files - where the differences between dithered and non-dithered is apparent - and see if I can hear differences between dither types.

BTW, do you simply skip adding dither (it is an extra step) or do you just go ahead and dither as it is easy and free?

Quote:
I am always tempted to reply that if you can't hear the differences between various forms of dither you need not worry about it.

I've replied to those a few times. Usually I ask "Why don't you just try it and decide for yourself? Sheesh!"

Quote:
do you simply skip adding dither (it is an extra step) or do you just go ahead and dither as it is easy and free?

In truth, I never bother with dither. Yes, it's free. But the kind of music I write never goes below -30 or so. So just on principle I refuse to waste 0.0003 cents on the extra electricity needed for the render to take 1/8 second longer.

Ethan Winer
RealTraps

Quote:
A file truncated from 24-bit to 16-bit without the prior application of dither contains artifacts that we can measure, but at such low levels it is masked other than in isolated circumstances.

Yet...read my experience with inadvertently truncating 20-bit data to 16 bits at http://www.stereophile.com/asweseeit/523/ . (Incidentally, as I was not aware of there being any reason for an audible difference, I was in effect performing a blind test on myself.) Also, see my essy at http://www.stereophile.com/asweseeit/372/ , where I detected that the data on a disc I was producing had been converted by someone else to analog and back to digital purely by listening (then verified by subsequent measurement).

John Atkinson
Editor, Stereophile

Thanks, John.

I just read the second example (where the ME changed your data for the Concert CD).

In this case the ME either "a) used a sample-rate converter, or b) converted my carefully prepared data to analog, then reconverted it to digital using a 16-bit ADC with a sample clock running very slightly slower than mine" (the lower clock speed was a great catch, by the way!).

These could indeed be audible and are not nearly as subtle as whether or not dither was added. Although I note that prior to sending the data off you tried the various dither methods and your disposal. Neat!

(Did you ever learn what he did? And why in the world he did it when all you wanted was the PQ subcodes?)

The other article (involving one of your wonderful recordings of Robert Silverman) provides examples of hearing differences between a 20-bit recording and a 16-bit truncated version. You also describe hearing differences between high and low jitter versions of the same recording.

Since this article also does not address dither, assume your point is that there are small differences that can matter and be audible. Very relevant.

I wish I had the opportunity to have listened to the various versions you describe.

It would be very interesting to null the high jitter v low jitter recordings, and the 16-bit v. 20-bit versions to see what results as the differences.

One of your examples of what you can hear between 16-bit and 20-bit is the end of reverberation tails. I have heard this difference as well.

Very low level signals are indeed better captured in my recordings with higher bit depths, especially since I record aiming to have peaks no higher than -18dBFS and thus like to take the extra bits as extra cushioning for the low level stuff.

It may well be that low level harmonics and the like are also better captured with greater bit-depth, leading to better timbre accuracy.

Edit: After all this I forgot to ask,

Can you hear whether or not the average recording has been dithered prior to bit-depth truncation? That is, can you reliably hear the difference between a 16-bit file dithered and that same 16-bit file without dither?

If you can, can you do so on main program material (not the end of long reverberation tails for example) at standard listening volume?

And if so, can you send me copies of the files?

Seriously.

The only examples I have ever heard where people claimed they could hear dither were extreme examples, such as 8-bit files and during incredibly low level passages listened to at very high volume.

It may well be that - like many things in audio and recording - that the artifact is hard to hear at first but once you identify it it is easy to pick out.

Quote:

Quote:
A file truncated from 24-bit to 16-bit without the prior application of dither contains artifacts that we can measure, but at such low levels it is masked other than in isolated circumstances.

Yet...read my experience with inadvertently truncating 20-bit data to 16 bits at http://www.stereophile.com/asweseeit/523/ . (Incidentally, as I was not aware of there being any reason for an audible difference, I was in effect performing a blind test on myself.) Also, see my essy at http://www.stereophile.com/asweseeit/372/ , where I detected that the data on a disc I was producing had been converted by someone else to analog and back to digital purely by listening (then verified by subsequent measurement).

John Atkinson
Editor, Stereophile

One consideration, as I see it, is that when recording live we don't know what the max level will be so the levels have to be set with some margin against overload. 20 bit and higher systems have a significant advantage in this area. In the "truncating 20-bit data to 16 bits" example were you recording with 6 - 24 dB of margin for headroom? This would represent allocating 1-4 bits for headroom, then if you truncate to 16 bits without normalizing, it would really be 15 to 12 bits actually used compared to normalizing. It is identical to recording to 16 bits setting the levels the same, but when recording at 20 bits or higher and distributing at 16 bits normalizing in order to use the full dynamic range of 16 bits is the way to go, IMO.

I do think John, that piano is a good test (one of many) for a digital system, and that it is possible to come to different conclusions depending on the source being recorded.

Pete Basel

Good point, Pete.

When recording at 20-bits one needs to be bit careful with levels.

I assume John dithered and changed bit-depth as his last step after settling final levels and SRC.

Quote:

Quote:
this is the video version of Ethan Winer's AES presentation.

Geared more towards the recording engineer, but some here may find it interesting. I know I did.

http://www.youtube.com/watch?v=BYTlN6wjcvQ

Hmm, my ghu, who was the ugly old guy who did the first talk after the intro?

I'm sort of annoyed that all the meat is removed, but so far, the editing hasn't changed the message, it's just removed all of the "why this happens".

I'd like to watch the rest of the 2 hours - some of us want to see the more technical parts.

Quote:
I'd like to watch the rest of the 2 hours - some of us want to see the more technical parts.

I have it all on video. Since you live near me, you're welcome to visit any time and watch it.

--Ethan

Quote:

Quote:
I'd like to watch the rest of the 2 hours - some of us want to see the more technical parts.

I have it all on video. Since you live near me, you're welcome to visit any time and watch it.

--Ethan

It was a great workshop. I would like to see the rest of the speeches, too. Is there any reason why the whole thing could not be put on youtube?

Just seeing this now. This video rocked my world. Not because of the discussion at AES, but because I used to play cello in quartets with Poppy Crum at chamber music camp!!!

Small world

Quote:
In the "truncating 20-bit data to 16 bits" example were you recording with 6 - 24 dB of margin for headroom? This would represent allocating 1-4 bits for headroom, then if you truncate to 16 bits without normalizing, it would really be 15 to 12 bits actually used compared to normalizing.

A good point. If I remember correctly, the piano recording peaked around -3dBFS. I tend to record with instantaneous peaks getting as close as possible to 0dBFS when recording, in order to maximize resolution. (While A/D converters don't perform quite as well as you get close to the maximum level, any additional distortion will only affect the half-cycle of the peak.)

John Atkinson
Editor, Stereophile

From the video:

"If the frequency response is flat to less than 0.1dB 20hz-20khz, and the sum of all noise and distortion is less than -100dB, a device will sound the same as any other audibly transparent device"

How many hifi components achieve this?

Quote:
A good point. If I remember correctly, the piano recording peaked around -3dBFS. I tend to record with instantaneous peaks getting as close as possible to 0dBFS when recording, in order to maximize resolution. (While A/D converters don't perform quite as well as you get close to the maximum level, any additional distortion will only affect the half-cycle of the peak.)

John Atkinson
Editor, Stereophile

John, do you do this when recording 24-bit as well?

Given that -18dBFS is roughly equivalent to 0 dBu (and represents only 3 bits) it makes sense to me to gain stage everything in a digital system as if it was analog.

I remember in the old cassette days when every graph of extended recording response was shown at -10db and -20 db levels to get past 15k, but who could do that with a -60 to -55 db noise floor. Most every tape you could buy had overs of +3 and +6db or more to maximize s/n.

I read some engineer felt that one you turned on the last bit at the -6db level, higher didn't matter. I know JA has said that in digital overs can be very bad, or something to that effect.

I am not a pro, but even if I normalized I kept my top at -1db. I did not want to go over. It just made some sense to have all the tracks at some even level of loudness. I never used compression and won't. Even on the crummy recordings I do.

Quote:

Quote:
A good point. If I remember correctly, the piano recording peaked around -3dBFS. I tend to record with instantaneous peaks getting as close as possible to 0dBFS when recording, in order to maximize resolution. (While A/D converters don't perform quite as well as you get close to the maximum level, any additional distortion will only affect the half-cycle of the peak.)

John, do you do this when recording 24-bit as well?

All my recordings are 24-bit. I don't understand why, when all real-world 24-bit ADCs are limited 1t 18-120 bits of real resolution, you would throw away some of that resolution and suffer a higher electronic noisefloor than necessary at the outset.

Quote:
Given that -18dBFS is roughly equivalent to 0 dBu (and represents only 3 bits) it makes sense to me to gain stage everything in a digital system as if it was analog.

I have never understood that concept, though I note Paul Frindle's gearslutz posting refers back to the days of analog and having a calibrated 0dBVU. I record all tracks as hot as possible (without clipping of course). Then when I mix by applying attenuation to each track, I am also attenuating the converter's own noise and distortion. Typically, if the converters offer 20-bit performance (and my dCS 904s do) and I am mixing in a solo instrument recorded at close to full-scale at -18dBFS, I am actually working with close to 23-bit resolution data.

My goal is to capture the maximum possible resolution at the tracking sessions, then do every mathematical operation in the mix with as much resolution as possible. Then, when I decimate and dither the final 24-bit rendered file to 16 bits for the CD master, I will preserve as much of the original resolution as is possible.

It has been argued that why bother, when the sources of analog noise ahead of the converter are higher than the converter's own noise? Having looked at this in depth over the years, "room tone" tends to have a "red" spectrum, ie, is gone by the time you reach the low treble where the ear is most sensitive and that is where you need more than 16-bit resolution.

And remember that I have total control of the project, from tracking through to mastering, and including setting the playback system characteristics, so I am free to apply a different philosophy to that practiced in commercial studios.

No one has yet complained about the quality of the CDs I have produced, at least in terms of resolution. :-)

John Atkinson
Editor, Stereophil;e

Quote:
No one has yet complained about the quality of the CDs I have produced, at least in terms of resolution.

All of your work that I have heard has superb sound. This is the primary reason why I ask.

Do you leave any headroom at any stage to protect against inter-sample digital overs?

I find that if I get eyeball RMS at approximately -18dBFS and keep metered peaks to no higher than -4dBFS or so that everything works out well. I don't trust digital "VU" meters to tell me where the absolute peaks are on the fly.

Quote:

Quote:
No one has yet complained about the quality of the CDs I have produced, at least in terms of resolution.

All of your work that I have heard has superb sound. This is the primary reason why I ask.

Thank you. But again I empathize that this way of working is only possible when the same person has control over every aspect of the project: trackng, overdubbing, mixing, and mastering.

Quote:
Do you leave any headroom at any stage to protect against inter-sample digital overs?

Oh yes. When I am mixing, I aim at the rendered file peaking around -3dBFS, which I then can normalize as wished. But occasionally I do get caught out on a hot peak, in which case I just readjust the peak section rather than redo the entire mix.

I mix using Adobe Audition 3, BTW, which does display the waveform assuming the effect of the reconstruction filter. This is most useful, because you can get a between-samples "over" from the filter's behavior even when the sample values are below full-scale.

This wasn't the case with the outboard Dorrough meter I was using with my original Sonic Solutions system - there was a single over for this reason on my 1997 "Rhapsody" CD that bugged the heck out of me (though I couldn't hear it, which is why it got on to the CD). When we repressed the CD last year, I took advantage of the opportunity to remaster it and fix the over. But no-one had ever noticed or complained about it. :-)

Quote:
I find that if I get eyeball RMS at approximately -18dBFS and keep metered peaks to no higher than -4dBFS or so that everything works out well. I don't trust digital "VU" meters to tell me where the absolute peaks are on the fly.

Broadcast purposes aside, I hate and despite VU meters. All they tell you is loudness and our ears are pretty good at doing that. Recording digitally, what you _need_ is an instantaneous headroom alert. Your practice sounds good, though I would add that it does depend on the music being recorded. Even if you tell a drummer to play as loud as he can when setting levels, he will always play up to 6dB louder on the night. Fortunately, a single clipped snare drum transient doesn't sound appreciably different from an unclipped one, and the pencil tool in Bias Peak does occasionally see some action.

The main problem with a digital over (provided a positive peak doesn't wrap over to a single full-scale negative sample, which produces a very audible "click" - ugh) is that the converter goes "deaf" for the duration of the duration of the clip. Not a big problem for a single transient; big problem with a succession of clipped peaks.

John Atkinson
Editor, Stereophile

Quote:
The main problem with a digital over (provided a positive peak doesn't wrap over to a single full-scale negative sample, which produces a very audible "click" - ugh) is that the converter goes "deaf" for the duration of the duration of the clip. Not a big problem for a single transient; big problem with a succession of clipped peaks.

John Atkinson
Editor, Stereophile

Sorry to disagree, but while that's true, the real truth is even worse.

When you clip digitally, or in a convertor AFTER the antialiasing filter, all of the harmonics and distortion you add that is ABOVE half the sampling rate aliases down to below.

This leads to really, truly, fantastically ugly distortion artifacts that are not harmonically related.

Needless to say this supports the idea of "no digital overs or else, please" rather well.

Hmm. I have a picture of that up somewhere, I think, let me look.

Nope, not on line. I could send you the deck with the slides in it if you like.

I've wanted a Dorrough meter for quite while, they seem so "pro".

Thanks for the info. Very helpful.

Quote:

Quote:
The main problem with a digital over (provided a positive peak doesn't wrap over to a single full-scale negative sample, which produces a very audible "click" - ugh) is that the converter goes "deaf" for the duration of the duration of the clip. Not a big problem for a single transient; big problem with a succession of clipped peaks.

Sorry to disagree, but while that's true, the real truth is even worse.

When you clip digitally, or in a convertor AFTER the antialiasing filter, all of the harmonics and distortion you add that is ABOVE half the sampling rate aliases down to below.

Thanks JJ. That's pretty much what I meant by "big problem with a succession of clipped peaks." But while the harmonics and other spuriae are required to square the waveform with a _single_ clipped peak, those disappear when the transient is over, which works against their audibility.

I tried this with the starting transient of a single snare-drum stroke a while back, starting with an unclipped capture, then stepwise increasing the mike preamp gain, to increasingly drive the ADC into flat-top clipping. (Not sample-wrap type overs.) Adjusting the playback level accordingly, I couldn't hear any difference between pure and clipped versions until I had chopped off, if I recall correctly, around 10dB of the single waveform peak.

So while yes, never clipping your ADCs is the best thing, accidents can happen, especially when you're recording live. Whether or not your recording is irretrievably damaged will depend on the nature of the overs. A single event - you're probably okay and you can redraw the waveform to minimize the nasties. Repeated waveform clipping, however - you're going to have to do some cut'n'pasting from an undamaged part of the recording if you don't have another unclipped take available!

John Atkinson
Editor, Stereophile

Quote:
I've wanted a Dorrough meter for quite while, they seem so "pro".

Thanks for the info. Very helpful.

My Dorrough dates back to 1993. I am sure that nore recent versions calculate the reconstructed inter-sample peaks rather than just showing the sample values.

John Atkinson
Editor, Stereophile

Quote:
Whether or not your recording is irretrievably damaged will depend on the nature of the overs.

More to the point, clipping a transient (something with a wide spectrum) is not going to create a total disaster.

Clipping something more stationary, that has some tonal content, on the other hand, is going to be very, very unpleasant.

If you think about it, distortion products for an impulse have the same spectrum as the impulse, ergo there isn't a big change in spectrum, just in level.

Now, when you have some tonal content you're in trouble.

Fortunately, that's not a common case, well except that it is on quite a few modern CD's. Finding a track where -32868 and +32767 are the two most likely levels (bin by bin histogram of the entire track) says a lot, and nothing good. In any kind of music, log-normal distribution is pretty good to assume, except when it's clipped. When you see soemthing where the min and max levels are about 1.5 orders of magnitude more likely than zero . . .

I don't think you need the dots filled in. Ask me privately if you want to know what track that was, and no, it wasn't Metallica.

The track, however, is (*&(* loud.

http://i238.photobucket.com/albums/ff228/jj_0001/plots/ughl.jpg

That's the histogram of the particular track. Yes, it's real.

Btw, I want you to hear these some day...

http://i238.photobucket.com/albums/ff228/jj_0001/plots/DSC_0003csc.jpg

Interesting stuff.

I know they have digital error correction, is there such a thing as digital clipping correction or progressive limiting?

Quote:
Interesting stuff.

I know they have digital error correction, is there such a thing as digital clipping correction or progressive limiting?

Not after the fact . You've lost information, and that's that. You can sorta, kinda mitigate short-term clipping. Kinda, barely.

But it's still in the "questionable practice" catagory.

Massive clipping means massive information loss. Gone is gone.

Quote:
That's the histogram of the particular track. Yes, it's real.

Wow.

Re digital reconstruction: My experience is that it can take some of the edge off of the distortion, but that's it.

A single clipped waveform that is causing a brief click can be redrawn or otherwise changed in an essential way to remove the clip but all you really are doing is hiding the click.

Thanks, j_j for your great additions.

Fun thread to read while waiting for the sample rate converter to do its job from this evening's recording of a brass quintet.

And no overs!

Quote:
Is there any reason why the whole thing could not be put on youtube?

When I asked the AES workshop chair if I could bring video gear and record it, at first he said no. Then a week later he said okay, as long as it doesn't end up on the web. I think it has more to do with union rules than the fact that the AES sells audio files of their workshops.

Also, Poppy Crum told me some parts of her presentation are proprietary, and said she'd tell me by email which parts were okay to use. But when I emailed her she never replied and I didn't pursue it.

On top of all that, and with apologies to JJ and Poppy, my interest was mainly in the common-sense aspects of why people think they hear stuff that in truth is not likely audible. JJ and Poppy both gave very detailed and highly academic talks. So I chose the parts I thought had the broadest appeal.

Plus I wanted to keep it to less than an hour.

--Ethan

Quote:
How many hifi components achieve this?

Most electronic devices can do that easily, even the cheap stuff.

--Ethan

Quote:

Sampling rate convertor ...

Whrrrr ... the sound of a CPU fan running on max.

Quote:
someone postulated that one of the big reasons the ITB do not sound as good as OTB mixes is that engineers are going way too hot with the levels.

That someone would be wrong, especially when talking about modern DAW software that uses 32 bits floating point internally. There was a huge thread about this at the Home Recording forum. Here's the link:

http://homerecording.com/bbs/showthread.php?t=295994

Of course I had to pipe up and explain why they were all wrong. If you follow it to the end, or just jump to the last page, you'll see that after all the insults, common sense eventually prevailed.

--Ethan

Quote:

Quote:

Sampling rate convertor ...

Whrrrr ... the sound of a CPU fan running on max.

You have this right! iZotope's 64-bit SRC at the highest quality settings keeps all four cores busy for quite a while each time I run it.

Quote:
I find that I get best results(sonically speaking)keeping peaks at -10 to -12. I work in the digital domain in pretty much the same manner as I do in analogue.

Exactly.

I run few plug-ins so I don't have much exposure to the potential issues discussed in the thread, but I follow the advice of watching gain staging carefully.

Besides, as you point out, Paul Frindle knows his stuff

Quote:

Quote:
I find that I get best results(sonically speaking)keeping peaks at -10 to -12. I work in the digital domain in pretty much the same manner as I do in analogue.

Exactly.

I run few plug-ins so I don't have much exposure to the potential issues discussed in the thread, but I follow the advice of watching gain staging carefully.

Besides, as you point out, Paul Frindle knows his stuff

Hunh.

Not sure what I did to make this disappear the first time, but in fact 24 bit mantissa is enough for about a 1024 point FFT/IFFT if you want to keep 16 bits. You may also have problems with sharp IIR filters, or long FIR filters in single-precision float, if you want to keep 16 bit resolution.

Not quite following you there, j_j.

As I mentioned earlier, I eyeball RMS at approximately -18dBFS and keep metered peaks (digital VU) to no higher than -4dBFS. If I see anything higher than this I turn the mic gain down the next chance I get.

Part of the reason is that I usually am recording a live performance on location and need to keep a safety net. Another reason is that the VU meters I have are not going to show me quick transients or intersample overs.

Are you opining that this is a bad idea or that I should normalize first before doing anything in my DAW?

Quote:
Not quite following you there, j_j.

As I mentioned earlier, I eyeball RMS at approximately -18dBFS and keep metered peaks (digital VU) to no higher than -4dBFS. If I see anything higher than this I turn the mic gain down the next chance I get.

Part of the reason is that I usually am recording a live performance on location and need to keep a safety net. Another reason is that the VU meters I have are not going to show me quick transients or intersample overs.

Are you opining that this is a bad idea or that I should normalize first before doing anything in my DAW?

I'm talking about floating point processing. This has nothing at all to do with having 24 bit input a few bits down. Sorry to confuse.

Quote:

Quote:
How many hifi components achieve this?

Most electronic devices can do that easily, even the cheap stuff.

--Ethan

I can't even remember the last time I saw a set of measurements from any component that achieved this.

Can you show me some?

Even a few feet of lamp cord as speaker cable like you recommend doesn't achieve it!

Quote:

Quote:

Quote:
From the video:

"If the frequency response is flat to less than 0.1dB 20hz-20khz, and the sum of all noise and distortion is less than -100dB, a device will sound the same as any other audibly transparent device"

How many hifi components achieve this?

Most electronic devices can do that easily, even the cheap stuff.
--Ethan

I can't even remember the last time I saw a set of measurements from any component that achieved this!

Me neither. (Or should that be "me either"?) There are a couple of very expensive power amplifiers and preamplifiers that achieve that goal, but that's about it.

John Atkinson
Editor, Stereophile