Holm Acoustics DSPre1 Digital Preamp/Room/Speaker Correction Device (TAS 208)
Digital signal processing offers amazing possibilities for improving the sound of an audio system. And the Holm DSPre1 really delivers the goods. It offers the ease of an automatic system together with the superior results that detailed user control allow. Purely automatic systems, for all their ease of operation, often leave one frustrated by the impossibility of fine adjustments. Enter the Holm. Its operation and programming are so straightforward and so easily comprehended at first encounter that a novice with DSP will easily and immediately obtain gratifying results. It is as simple to set up as the purely automatic systems. But the control offered to the user, again in an easily approachable form, means one can do really fine adjustments by ear—and much for the better. A quick look at the instruction videos on the Holm Web site and you will feel like an expert—and you will get expert results as well.
For those who want to go deeper, there are many further options, reaching all the way to being a do-it-yourself crossover designer, enabling you to make DSP crossovers for any bi-ampable (or tri-ampable) speaker. In addition, its intrinsic digital sound quality is of the highest level. The Holm DSPre1 is superb simply as a way to play digital sources, leaving the correction systems aside. And it allows corrections to be made that will take the sound of your system to almost unimagined perfection. It is truly a DSP system for all, from the novice to the professional and everyone in between, and a sonic wonder.
There are important features of the DSre1 that distinguish it from previous DSP units. For one thing, it verifies before your very eyes by before-and-after measurements that it did what it was supposed to do. DSP is, in effect, mathematics, and mathematics can be taken on faith—it always works. But hardware has limitations, and it is useful and important to be able to check that an implementation of mathematics in hardware form actually did what it was supposed to. An even more important distinctive feature of the DSPre1 is that it does what many people including me have long considered to be the right thing to do—it corrects the speaker first and then corrects its interaction with the room afterwards. Why this is a good idea I shall go into later. But it definitely is a good idea, and it works out extremely well here.
Another important feature of the DSPre1 is that it operates on digital directly. Some otherwise admirable DSP correction systems offer only analog in and out to do the DSP. With these, you have to convert a digital source to analog, then have the correction device convert the analog back to digital, and then, after the DSP, convert back to analog. While the DSPre1 will accept analog inputs and handle them very well, it also accepts digital inputs and operates on the digital signal directly. No extra conversions are involved.
How It Operates
Like all DSP correction devices, the Holm DSPre1 performs measurements of your audio system first to determine what corrections to do later, when you play music. The Holm‘s initial measuring and programming of DSP corrections is computer-controlled (USB connected) by a Windows point-and-click program. The instructional videos from the Holm Web site get you up and running, though the screens presented by the device itself are so clearly laid out that you could probably dispense with the videos if you wanted to. A calibrated microphone is supplied. Attach the unit to a line-level preamp input or, better, directly to an amplifier, attach the mike to the unit, attach your computer with the program installed to the unit by a USB cord, and you are ready to go. When the measuring is done and the system optimization is over, you can disconnect your computer. But you can also leave the computer connected to change EQ curves on the fly—or to play music from your computer.
The DSPre1 functions as a digital line-level preamp; it includes a volume control so that its analog output can be connected directly to an amplifier. The DSPre1 accepts external digital sources of a wide variety of sampling rates as inputs (see specs at the end of the article). It can also accept analog inputs (line-level), which are A-to-D converted internally so that DSP can be done on them, and these also admit volume adjustment. A separate analog preamp is not needed. Alternatively, one could run the output into an analog preamp line-level input, if one wanted, for example, to have a phono setup that was not run through the digital processing—though once you have heard the improvement the DSP makes, I doubt you will want to give it up.
Measurement for room and speaker correction is a two-stage process. And here we get to something distinctively Holm: the separate measurement of the speakers themselves. First, one puts the microphone supplied—which has a precision calibration curve that the unit uses to make the corrections correct—near the left speaker, on the listening axis but closer than the listening position. The unit checks volume levels and then runs a “log sweep” frequency-response measurement of the left speaker. (Watch out! Loud high frequencies here, so acoustic isolation headphones or earplugs are recommended). Log sweep measurements seem all the rage in Europe for some reason, rather than the chirps or max length sequence signals that have been popular in the U.S. for some time. But mathematically, it all comes out the same. The idea here is that one can get something essentially like the anechoic measurement of the speaker from around 300Hz on up. The unit then corrects the speaker in the sense of designing a signal alteration that will make the correction plus speaker have flat response.
This works in the usual way on amplitude response: If there is, say, too much at 3kHz, then the correction pulls down 3kHz and so on. But in addition to correcting frequency response, the DSPre1 corrects phase response. So now your corrected speaker is flat and also phase linear—even though it was likely not phase linear before the correction. (Speakers with analog crossovers of a higher order than the first are never phase linear—the crossover generates phase nonlinearities in the total output of the speaker. When people say that the popular Linkwitz-Riley fourth-order crossovers are “in phase at the crossover” this does not mean that the output is phase linear, it just means that the drivers are in phase right at the crossover. There is in fact a 360 degree phase rotation over the whole range with such a crossover).
The unit then re-measures the left speaker. With the mike left where it was to start with, the unit does its log sweep once more and shows you how flat and phase linear the speaker is now. And of course flat and phase linear make the impulse response really good. The unit shows you that, too. What you see is definitely what you get here. Then you move on to the right speaker, putting the mike close and on axis with it, and do the process again.
About this impressive process there is one caution: You have to have the microphone on the axis that has the least cancellation between the drivers. If you put the mike in a position where there is a dip caused by interference between the drivers, the correction system will insert a peak to compensate—and you won’t like the results in listening. Holm suggests experimenting with positioning the mike when doing the speaker corrections. Good advice! Run the measurement several times with the mike at different vertical positions near the tweeter axis until you find the one that has the fewest dips. Then use that one—but, of course, make the same choice for both speakers! This is really important. If you do this wrong, things are not going to sound good later on.
Now, with your speakers all fixed up, you take the mike back to your listening vicinity, further from the speakers. Now the system measures for room effects. This measurement is done over several positions successively: the main listening position, to the left of that, to the right of that, above it and in front of it. The point of this measurement over space is that one does not want to correct highly localized effects but to get something that works over a “sound cube” around the listening position. The DSPre1 plus computer then develops a correction algorithm automatically and quickly and stores it in the unit to use when music is being played.
Now you have a basically flat system. But in addition, you have choices of the final result, choices for the sound you will actually hear. First, there are some standard curves that differ slightly from flat but that give pleasing sonic results, reminiscent of things like the BBC dip and so on. In the brighter of my two rooms, I liked the one called Deep Tilt II, which pulls the bass up slightly and does some relaxation in the treble, but in my softer room, I also felt like the unvarnished truth of flat EQ on some material. EQ is easy to change. You can in fact change it while the music plays, if you leave the computer attached.
You can also construct a custom curve and modify the balance of the sound in any way you like! This is easy to do, too. A simple point, click, and drag system on screen enable you to draw any curve you like, starting with the flat curve (or one of the alternatives explicitly offered). Want a CD to sound like a moving coil cartridge playing a record? Put in a little bass bump, a presence range dip, and a flip up of the top end. Want to tame a bright recording? Drag down 8kHz and get relief. Want to rock out? You can drag the middle bass up in a few moments. (Watch out for possibility of woofer overload though!). And so it goes. You are in charge in a big way. And it is all very easy to do, right on the spot as the music plays if you wish, in “real time.”
Maybe this will make some people nervous—to have so much control and be an active participant in what you are hearing, not a passive victim. But audio is like existentialism: You always are making a choice of equalization, no matter what you do. You cannot escape the existential responsibility. Letting the system you bought do what it does is still choosing. It is just a choice you cannot modify unless you buy some other equipment, which you again accept passively. But here you can change your mind—even from one recording to another. Or you can find a setting you like and just live with it. In any case, remember: You have an EQ’d system. The only question is whether you are choosing the EQ yourself or let the manufacturers of your equipment and the nature of your listening room choose for you.
One of the things you will learn fast, if you do not know it already, is that even small changes in frequency response make big changes in the sound heard. Those graphs you see in magazines sometimes that make things look flat by using a compressed vertical scale, with 5dB being an eighth of an inch or the like, turn out to be really deceptive. With the threshold for audibility of broadband shifts of response being on the order of 0.1 and 0.5 dB, you can see that being flat in, say, +/- 3dB terms does not really cut much ice towards really determining the sound heard. [The psychoacoustic literature suggests that the threshold of audibility of frequency response shifts is 0.1dB over an octave of bandwidth. —RH] Being able to move response curves around easily and on the fly, with just a few seconds of reloading filters to go from one to another, really brings this home. Exactly which response curve you like the best may depend on the material and on room acoustics and speaker radiation pattern. But for sure, you will find that small changes in the curves you choose really matter. The idea that it was all taken care of already, that your system already was neutral in a definitive way, won’t survive long.
Why Correct the Speakers Separately
One of the important aspects of the Holm DSPre1 is the correcting of the speaker itself, independently of room effects, before undertaking the room correction. It is worth a close look at why this is a good idea because the reason is not obvious. After all, EQ is EQ. As far as gain is concerned, and phase, too, there is one final curve, in effect. The mathematical concatenation of filters is just another filter. Why not do it all at once by just measuring where you are going to be listening?
The reason for treating the speaker itself first is psychoacoustic. When one listens to a sound source, in a room or not, the ear/brain is constantly interpreting the changes that occur with head movements. In effect, you are trying, without knowing you are trying, to hear what the source sounds like as a source unto itself. You are trying to separate out room effects.
You can see that this works. It is why when you listen to someone walking around a room and talking—say, a lecturer walking back and forth in front of a blackboard—the sound of the voice does not change even though the literal physical sound you are hearing changes quite a lot. But the ear/brain edits that out completely. Well, almost completely. Actually if you listen hard, you will in fact hear the voice get more bass-oriented if the speaker gets close to a boundary or, even more, close to the corner of the room.
This illustrates the whole thing, actually. The ear/brain tries to and largely succeeds in editing out room effects in the high frequencies, but in the lower frequencies the room and the source are mixed together. And of course in the middle frequencies there is a transition. Roughly, the good model is that there is a time window which is longer at low frequencies and shorter at high frequencies and what happens in this window is what the ear/brain hears as the real sound.
But there is that additional dimension that the ear/brain keeps track of things using head movements. What this means is that the ear/brain has access to information that depends on spatial variation. For example, a tiny peak in frequency response of a loudspeaker that is the same over an area will be detected in a way that a tiny peak that occurs on account of a room reflection and that varies a lot with head movements is not detected. Floyd Toole once summarized this by saying: Average over space, not over frequency.
If one uses a short time window with a microphone at the listening position, short enough to get direct arrival only, this in effect does frequency domain smoothing whether you want it to or not! This is an aspect of Heisenberg’s Uncertainty Principle actually. A short time window does not allow high resolution of one frequency from another. Of course, mathematics applies to the ear, too! But the difference is that the ear/brain can and does average over head movements, which a microphone cannot do. (One could, in principle, simulate the ear/brain spatial averaging, but it would be complicated to do and the model of exactly what happens is not really known all that precisely.)
What all this psychoacoustic information (call it babble if you wish, but it is really true!) amounts to is that one can do better in figuring out what the ear/brain will actually hear if you know what the speaker is doing anechoically.
But, you might say, are speakers not smooth and flat by design? You wish! You can see visually that speakers have spatially persistent irregularities, almost all of them, by looking at the NRC measurements of a great many speakers available on www.soundstage.com. Those little blips up and down—and virtually all speakers have quite audible ones—that persist over a variety of off-axis measurements are important. And most of the little on-axis blips, in fact, do persist over space quite a bit. Designers are a lot better at getting rid of early reflections off the speaker itself and diffraction effects, the things that make blips that vary in space, than they are in making the speaker actually smooth and flat in the spatially invariant sense. Most of the errors on axis also appear in the near off-axis. So you can get a long way by correcting the on-axis response. Maybe it would be better still to average a bit over small variations of position, but in practice if there is a blip at a certain frequency in the on-axis anechoic response, it is likely to be there just off-axis, too.
Hence, putting it all together (if eyes have not glazed over by now!), it is a good plan to fix the on-axis response as measured anechoically. The DSPre1 cannot make your room an anechoic chamber, but it can and does let you measure up pretty close. If you are really fanatic (and most of us are, I suppose), you might want to take your speakers out of doors. But it turns out that the differences are small in the higher frequencies between on-axis up close and on-axis out of doors. So the Holm method of up close in-room measurement works well.
After all these preliminaries of an operational and theoretical nature, curiosity must have arisen about the actual sound. First of all, the intrinsic sound of the digital here is absolutely top drawer. Holm has a jitter suppression method that seems to work superbly, and the overall sound of the Holm stack (DSPre1 and transport) is right at the top of the sonic heap in my view. In this sense, the Holm stack would justify its price versus the other high-end digital competition on its intrinsic sound alone, without even considering or using the correction processes.
But it is the correction processes that make the Holm the item of fascination that it is. Now a typical audio review usually involves a lot of description of frequency response variations, albeit often small ones that are heard as something else, as likely as not. Since you can make the frequency response of the Holm into anything you like, this type of review by frequency response is irrelevant here. The device is neither dull nor bright, bass heavy nor bass light, midrange forward nor midrange recessed. Such things and anything in between are at the disposal of the user!
Still, there is a characteristic sound to the corrected systems. First of all, the sound is extremely well behaved in the bass. Uncorrected systems, effectively all uncorrected systems, sound loose and muddy in the bass if one compares to the same systems corrected. This is not a question of lean sound, of definition obtained by bass deficiency. Rather the bass, even when it is augmented by one of the presets or by a (reasonable) custom target curve, has a firmness, smoothness, and solidity that escape uncorrected room/speaker setups. Things like tympani parts and other bottom-end transients in orchestral music acquire a precision that is surprising, but without losing their impact. The effect on the bottom end of the well-recorded Dvorak Legends [Philips], with Fischer conducting the Budapest Festival Orchestra, for example, was spectacular. Bass pizzicato notes acquired the pitch and definition of real life with real-life warmth intact. (In most non-DSP situations, bass definition is a code word for bass deficiency—the perceived definition comes from lack of real bass power and extension and consequent cancelling of masking effects—cf., any mini-monitor and its “transparency.” Not here!)
Second, the corrected systems have a remarkable smoothness of sound all the way up. Again on the Dvorak Legends, the soaring first violin parts had the beauty and liquidity of reality, and balance continued to be perfect all the way down. With the Holm, nothing stands out excessively or is suppressed unduly. One hears something that sounds like the smoothness of real music. To say this is neutral is to understate the case, given how people tend to throw the word “neutral” around nowadays. This is real neutrality in a sense that effectively no uncorrected system offers. This in turn vastly diminishes the sense that one is listening to speakers. A speaker made perfect is a speaker that no longer sounds like a speaker. You just hear music.
Third, stereo focus is startling. Even on a typical set-up CD with a mono signal in polarity and then out of polarity will show already the surprisingly exact focus of the in-polarity two channels and the total lack of focus (as it should be) of the out-of-polarity situation. You have to hear it to believe it. And of course this carries over to music. Instruments are really “there” in a way that they seldom are in reproduced music.
Finally, the phase correction makes a subtle but definite improvement in the realism of transients and attacks and in the definition of complex music. I have tried this often enough before, the comparison of phase-linear with not-phase-linear, cf. my review of the Arion/Essex phase and amplitude correction system, reprinted on www.regonaudio.com. But as years have passed, I have come to feel that this is even more significant than noted in that review from more than a decade ago. Phase linearity is not as obvious as frequency-response (amplitude response) effects, but it does count and make for enhanced naturalness.
The piano, for example, sounds to me considerably more like an actual piano when the shape of the initial transient is actually correct. Listening to something like the remarkable recording of Janne Mortensen playing Chopin (a demo disk from Gradient Loudspeakers) gave an absolutely minimized sense of discontinuity from the literal piano sound of the Steinway in my living room. And James Boyk’s superlatively recorded Tonalities of Emotion on Performance Recordings sounded uncannily realistic, even in direct comparison to the real thing. Not that the two concert pianos involved literally matched my own smaller grand as such, but the generic essential “piano-ness” was preserved in a way that seldom happens. And the bottom ends of the big piano were absolutely convincing, too.
The composite effect of all these improvements is remarkable. There have been over the years of high end so many audio reviews describing small changes (cables, say) as revelations that it has become hard to summon up rhetoric to cover something that really is a major change. But the effect of the Holm DSPre1 really does live up to those traditional and over-used descriptions: “revelatory,” “unprecedented realism,” “top to bottom transparency,” whatever your favorite buzzwords are for the sense of hearing the recordings in their entire truth.
In stereo performance and in intrinsic sound, in timbre and correctly perceived dynamics (which are closely related to frequency response on account of Fletcher Munson), in the overall sense of hearing not reproduced music but real music, the DSPre1 is a completely convincing illustration of DSP done right. And the wonders available are really only possible with DSP: No speaker is as flat and phase linear and in-room correct on its own as it will be after Holm correction.
People who know in-room measurements often express amazement at the in-room measured results for the Harbeth M40s in my listening room. Part of this is from a lot of work on careful placement and room treatment, and part of it is luck—the (pre-existing, not purpose-built) room just turned out to work really well. And of course the Harbeth M40s themselves are very flat indeed on their own and also are designed to integrate well into rooms. But even this room/speaker setup that works about as well as a system can on its own and works far better than many or even most, even this system sounded substantially better post-Holm. In particular, the lower frequencies were smoother and better defined, without any loss of warmth and fullness. The Holm really works—it will improve any system. And the better what you start with is, the further you will get towards perfection.
Learning What Happens
Isaiah Berlin used to separate the great thinkers of the world into two camps, the Hedgehogs and the Foxes. The Foxes know many small things and see the world as made up of a great collection of small pieces. The Hedgehogs see the world as dominated by a single great principle and interpret experience in the light of that one over-arching idea. Audiophiles tend to be Foxes: Having observed that small changes are audible, they tend to feel that perfecting an audio system is a matter of optimizing almost infinitely many small things. (Sometimes this can lead to really odd ideas. I have had audiophiles suggest to me that they are, say, going to try to correct a 10dB bass boom in their systems by getting different cables and the like). But experience with a device like the Holm DSPre1 tends to make one a Hedgehog. While the Holm unit itself is perfected in detail (e.g., in extreme minimization of jitter), the possibility of controlling phase and frequency response tends to demonstrate to an extreme the fact that a speaker’s behavior is largely determined by its behavior as a linear system.
People have known for some time that, as speakers have improved through improvements in drivers, distortion has gone down a great deal. And the BBC found out long ago how to push cabinet sound out of the range of audibility. This has not stopped high-enders from trying to push it even further down. But systematic tests have largely shown that the sound of a speaker is dominated by frequency response, radiation pattern, and phase response, by the behavior of the speaker and the room around it as a linear system.
There is nothing one can do by DSP of a speaker to change its radiation pattern (unless one goes for the active, bi-, or multi-amped speaker with DSP crossovers different from what the speaker had to begin with). Dipoles are different from forward radiators and sound different, wide dispersion sounds different from narrow, wide front baffle put the baffle-step at a different frequency from narrow fronts. (The wide front is better, in spite of current fashion: cf. link.)
Equally, only crossover revision and multi-amping will change distortion levels. EQ as such does not. These active speaker options are offered in the Holm unit (as additional price options)—I shall get to those in a Part II, to appear. But even sticking to just correction of the speaker as is, which leaves one stuck with its radiation pattern and its distortion behavior, a startling level of improvement is possible using the Holm unit.
The Hedgehog viewpoint in the Berlin sense would have it that with enough judicious adjustment, one can make speakers sound remarkably close to correct simply by adjusting them as linear systems, that is, by phase linearizing them and carefully adjusting the frequency response. And I think that this is true to a surprising extent. This idea is for me primarily a matter of practical experience—I have tried it on a lot of speaker/room combinations. The idea is also consistent with the “standard model” as I call it for the perceived timbre of speaker sound in rooms, to steal a phrase from elementary particle physics. (In this model—as developed in detail by Sigtech in the 1990s for example—the timbre one hears is dominated by energy content as a function of frequency, the content being measured in a time window that shrinks as frequency rises. The size of the window is to be taken as determined.).
This model is of course not quite total—other things have acoustic effects, and these effects can be important, especially in a live room. Power response and radiation pattern do count, and in a live room, they can count a great deal. One needs to start with a speaker that has a good radiation pattern that varies in a sensible way with frequency and also to start with a room that is “soft” acoustically. But within those limits, one can make the timbre of the speaker/room combination remarkably correct via equalization , provided that one has enough control over what EQ is used. The control is needed because the secondary effects arising from reflections and room sound mean that slight deviations and compensations from any standardized measurement and correction system are likely to be needed. The situation is so complex that an automatic system will likely need some tweaking by listening to sound its best.
The Foxes, the believers in the importance of every detail of a system, may find it hard to accept how well this process of correction by EQ can work, how close to ideal speakers can be made to sound by this one thing only. But experience is the great teacher. Try before you decide! To get nearly perfect, you need a good radiation pattern, low distortion, and a room that is soft enough. But there are speakers that have those, or close enough, and arranging this about your room is likely not too difficult (curtains, soft furniture). The speaker/room combination will then come up with proper EQ (including phase linearization) into a state of near perfection that will stun you if you have a vivid recollection of actual music. The precision of the bass, the smoothness of everything, the perfection of initial transients, the over-all naturalness of the sound, are stunning and extremely satisfying in musical terms.
If you listen often to real music, to the absolute sound in the original meaning of the phrase, the Holm DSPre1 will be a thrill at first listen and a source of deepening regard and pleasure as time passes. The Holm DSPre1 is not inexpensive, but it will do much more to alter the sound you hear than is usually the case with changing to expensive electronics, and it will give you levels of control over the sound of your system that no straight-wire-with-gain device even tries to equal. There is nothing like the feeling that you have got things right, and the Holm DSPre1 will provide access to this sensation to a truly remarkable extent. With the DSPre1, you can get it really right at last.
Specs & Pricing
Digital preamplification and room/speaker-correction device with two-way or three-way DSP crossover options
Jitter: Below 10 picoseconds
Inputs: USB1.1 (control and 44.1/16, 48/16 digital inputs); TosLink 11k-192k 24-bit; four S/PDIF 11-192k, 24-bit; two analog, 88.2/24 A-to-D
Outputs: Stereo analog, volume attenuation 0 to -32dB by analog internal current adjustment, -33dB to -70dB by digital multiplier
Price: $8500 (DSPre1, 1-way)
Price: $7300 (CD1-D transport)
KELLY AUDIO TECHNOLOGIES
4613 Mount Putman Avenue
San Diego, CA 92117, California, USA