Katz's Corner Episode 17: Perfecting Perfection

This story originally appeared at InnerFidelity.com

Continuous Improvement
Hi again, fellow headphone investigators! The article I had originally planned to be Episode #15 was shelved and replaced by #16: The Smoking Gun. Since then, the LCD-4s which I reviewed have been replaced by Audeze because at some point they were damaged—no one knows where or how.

Regarding the Focal controversy, I personally disagree with Tyll's decision to keep the Focal Utopias on the wall of fame, since no one seems to be able to find a current-production unit with the warmer, more neutral characteristics that had earned it a place on the wall in the first place. As well, I would like Audeze to send Tyll four or five current samples of the LCD-4, to confirm manufacturing consistency. But I do applaud his tentative placement of the LCD-4 on the wall. I've discussed the issue of Audeze's manufacturing with Tyll and with Sankar from Audeze and I feel confident that their quality control is excellent. They put every headphone through careful measurements in a test jig. Things just sometimes happen at the bleeding edge where we audiophiles tend to live. Every LCD-4 is practically custom made, there may not be any in inventory at a given moment.

Bob's Philosophy of Sonic Neutrality
Today I'd like to discuss what I mean by sonic perfection or sonic neutrality. Most audiophiles have observed that excellent playback systems of different manufacturers can sound radically different! Even the most accurate loudspeakers are subtly voiced and are different sounding. But how can there be more than one "right"? Philosophically speaking, two different-sounding playback systems can't both be "right": one of them must be correct and the other must be incorrect! Or maybe both are wrong!

Obviously there are systems which sound more "right" than others and no reproduction system is perfect. Plus, our tastes vary in loudspeakers and headphones. Some of us tolerate certain anomalies more than others. With correction, a loudspeaker can measure extremely flat, with the anomalies greatly reduced, but we have to use good, transparent correction or the cure can sound worse than the disease. And we have to understand how to interpret our measurements and know something about psychoacoustics. Many "automatic" correction systems sound horrid and far worse than no correction at all.

The Chicken and the Egg
We acknowledge that the stereo recording process is imperfect and recordings are only a semblance of reality. You can't fit a 100 piece orchestra in your living room. Nevertheless, we adapt, we can at least compare reproduced tonality to the sounds we know in real life.

We use recordings to judge our loudspeakers. We use loudspeakers to judge our recordings. That is the ultimate chicken and egg problem! Dr. Floyd Toole calls this "the circle of confusion." Recordings vary so much in tonality that it's difficult for us to decide which is the right recording to judge the correctness of a loudspeaker or headphone. And here's a rub: How can any reviewer make a judgment about a transducer without knowing what a recording is supposed to sound like? Some reviewers and consumers may leap to conclusions about a playback system without knowledge of what a recording is supposed to sound like.

As readers of audio reviews, we like to see the reviewer's reactions to a bunch of recordings that we are familiar with. The only problem is that the recordings we think we know have been colored by our memory of what they sound like on our own playback systems, and our own systems are inaccurate, in varying degrees! When we visit a friend's house and listen to recordings we are familiar with, we sometimes negatively react when it sounds different from our own reproduction system, but we neglect the fact that we tend to like what we are familiar with. What if our own familiar system is at fault and the other system is telling the truth?

You'd think that only recording engineers who were present at the session could judge a transducer. As an audiophile recording engineer for decades, I'm very familiar with the sound of my own best recordings, which I've heard on numerous playback systems. I know how I made these recordings, and where all the bodies are buried!1 I also use familiar recordings made by other engineers which sound excellent on my reference playback system, to avoid the "not invented here" syndrome. Being familiar with a recording does help, but it is not essential as we shall see.

The first step in identifying whether a system is accurate is to listen to a wide variety of recordings on the playback system, loudspeaker or headphone, and take very careful notes of the tonal anomalies you think you hear. If every single recording that you listen to produces a similar tonal anomaly, then it's probably safe to conclude that the transducer (or perhaps the power amplifier or converter or even the cables) is the cause of the issue. But what should you conclude if only some recordings produce the anomaly and others do not?

A lot of the time, recordings themselves are at fault. The bass range is particularly difficult to control and so recording and mixing engineers struggle with uneven bass in their control rooms as much as us music listeners. The classic case is the common popular music mixing practice of using a Yamaha NS-10 loudspeaker on top of a console: That's an acoustic nightmare! As a mastering engineer, I usually can tell a recording was mixed with desktop NS-10s by the heavy bass drum and the weak bass instrument. Often my job as mastering engineer is to fix problems caused by poor monitoring during the recording or mixing session. Many transducers are hyped at the high end to give a kind of "hi-fi impression" and a false sense of transparency. But sibilance and cymbals are not exaggerated in real life.

My best reference for tonal accuracy is of course the tonality of real life and the many acoustic recordings I and others have made that try to approach real life. I also subscribe to the principle that the most accurate loudspeaker or headphone system is the one which reproduces the widest set of reference recordings without requiring EQ. That's how I came up with this bell curve (Fig 17-1):

KatzCorner_Ep17_Graph_BellCurve

Fig 17-1: Bob's Bell Curve of Transducer Accuracy

A loudspeaker or headphone that only sounds good with a small subset of recordings or that satisfies fewer listeners must be less accurate than one which successfully reproduces a greater portion of reference recordings. Audio expert Dr. Floyd Toole has demonstrated that averaging all the different consumer speakers (some bright, some with too much bass or midrange, etc.) one ends up with a very flat curve, which is empirical proof that mastering with an extremely accurate and flat playback system yields a product that sounds correct on more systems. So if you want the best translation, pick a loudspeaker with the flattest response. Likewise, if you are looking for accuracy, pick the headphone that successfully reproduces the widest variety of good recordings.

Unlike loudspeakers, headphones should not measure flat and so we use a correction curve like the Harman curve promoted by Tyll. If we subtract the correction from the measured response we should get a flat line. Assuming that the correction curve is also correct! I use a modified curve, with slightly less bass boost than Tyll advises and I've tweaked the upper midrange/treble boost until the corrected headphones closely resemble my reference speakers. But I agree with the general shape proposed by Tyll and Sean Olive.

Another challenge is how to pick your reference recordings. I advise you pick a group of at least 20 fine recordings that you love, ranging from, say, brightest to dullest: always play these recordings when evaluating headphones. If your headphones reproduce a large quantity of reference recordings without exhibiting common tonal anomalies, then it's a fair shake to say they are accurate.

One recording may not have enough information in a particular frequency range to excite an anomaly. In episodes 13 and 14 we did nail down some clear anomalies with both the Focal and the Audeze, "family resemblances" that are common to a given headphone within a large set of good recordings. For example, I found some warm recordings which do not excite the sibilance range, so the sibilance anomalies in the Focal were not always excited. However, all six listeners felt a large portion of the reference recordings to sound far more sibilant on the Focals than on the Audeze or the loudspeakers. Conversely, the consensus of all six listeners is that the LCD-4 is probably boosting the sub bass a bit, warming up the lower midrange and is a little bit weak in the extreme treble. However, in upcoming episode 18, you'll find that the latest iteration of the LCD-4 is even more tonally neutral and does not have any obvious bass boost. I truly believe Audeze is trying to get to sonic neutrality.

Bob's Headphone Evaluation Method
To sum up, here's my approach to reviewing headphones:

a) If a large portion of a set of high quality recordings exhibits a certain anomaly when auditioning transducer #1 but not when auditioning transducer #2, then we can conclude transducer #1 is probably less accurate in that frequency range. The key here is to listen to a large number of good recordings in different genres and recording styles.

b) If the reference loudspeaker system does not reveal the anomaly on the recording in question, but it is audible on headphone #1 and not on headphone #2, then we conclude that headphone #1 is more accurate compared with the loudspeakers.

c) Adding some more expert listeners gives us a lot more data points, as in Episode 14.

So our best shot is to combine methods a and b, and if possible c. In addition, if we are very familiar with a recording and have heard it on a large number of systems, we can give increased weight to that recording. All this takes us closer to a very accurate subjective judgment of headphones' sonic neutrality.

You say tom-ay-toes
Recently a post at Head-case criticized my use of the terminology "lower treble" to describe the range circa 10 kHz. The fact is that there is no standardized vocabulary of defining a given frequency range. Where does "the upper midrange" lie? Is it 5 kHz-centered? That's where it is for me. For some, the "upper midrange" is around 2 kHz. While I try to be consistent in my terminology my dilemma is that there aren't enough terms to precisely describe all the frequency ranges that interest me.

How do you label the ranges 250-500 Hz, 1-2 kHz, 4-6 kHz, 8-10 kHz, and above 10 kHz? I like to call the frequencies above 13 kHz or so the "air frequencies", for example, because they are above the traditional high treble. Some people lump 4-9 kHz into one range that they may call "upper-upper midrange" or "lower treble". In my book, "Mastering Audio: The Art and the Science", I have a chart of frequency ranges and we discover that there are overlaps and differences in how people describe those ranges. So to be less ambiguous I ought to use the exact frequency range values in question. But I like to write enjoyable language so I sprinkle my vocabulary with words instead of numbers! Just remember that when we differ in the frequency language, neither of us is wrong.

If leaping to conclusions was an Olympic event, audiophiles would get the gold medal. I placed one little adjective in front of the name of the O2 amplifier that incensed another Head-Case participant, although my previous reviews made it clear how I place the O2 amp in the pantheon. (It's a lightweight. It gained its "fame" by semi-marketing more than by its sonic performance. But it is an inexpensive reference that many people know, and it was important to note that it failed to drive the LCD-4 with anything approaching good sound.) Another Head-Case participant criticized my discussion of the left-right Focal sample drivers being off in sensitivity by 1 dB. He said, "at least run a frequency sweep to verify." Frequency sweep is actually a poor way to judge left-right balance: Have you ever tried to evaluate 5 kHz sensitivity by ear using a sine wave? It's not a pleasant thought. Besides, individual frequencies sometimes shift with nodes that exist in the cans, but if the octave band is balanced in the center, then the balance is probably fine.

I've been around the block a few times and am well aware of these issues. If the complainer had read the entire paragraph he would notice that I did play mono pink noise. After level balancing, the pink noise imaged narrowly in the center at all main frequencies, certainly well enough for me to conclude that the two Focal drivers were well matched in level and frequency response after a simple level correction. So, folks, please be a little more tolerant—before you gossip in "the Internets", ask me about it. Trust first and verify. I always strive to be better; communication is better than criticizing someone on line.

Bob's Reference Loudspeakers and Room
In order to judge headphones against loudspeakers, we must use an accurate set of loudspeakers. In Studio A I master recordings which are going to be enjoyed by the public. My goal is to produce masters that sound good everywhere: meaning that they translate to the car, clubs, Hi-Fi, small systems, and audiophile systems. By inference, when my masters translate to a lot of playback systems, the more accurate my mastering loudspeaker.

The power to this room is a 100-amp dedicated service, divided into 5 20-amp 120 volt circuits, each home run back to the distribution/breaker box in the room. That forms a central point and minimizes loop area. The circuits are protected by Surge-X series protectors with EMI and RFI filtering. There are no MOVs on this circuit, which can contaminate the power by leakage current. I analyzed the line voltage with an FFT analyzer, line to neutral and line to ground. After the new power installation, the power line was about 20 dB quieter! This has resulted in a cleaner and quieter audio. Also, I now have more audio headroom: the dedicated AC power does not sag on peaks, and so delivers 4 dB more output level from the power amps before clipping compared to the regular house wiring! The main power amplifiers are two ATSAH 500 by Acoustic Imagery, each which deliver 700 watts continuous into 4 ohms and much more on short peaks. In addition to two 1500 watt subwoofer amplifiers.

My room has been trapped with both active and passive traps. The reverberation time versus frequency is very even. The loudspeakers have been corrected through FIR processing to produce an astoundingly linear frequency, phase and impulse response. This translates to a transparent, impacting, musical and accurate playback. Here's the frequency response curve of the left and right channel at the listening position (Fig 17-2), before and after correction, measured with Acourate software by Audio Vero. It's corrected to my preferred "target", which is flat from 20 Hz to 1 kHz, and a little more than 6 dB down at 20 kHz. This target sounds neither too bright nor too dull using a set of reference recordings. It helps deal with the natural rolloff with distance in the air. As you can see, compared to that target, my system is remarkably flat. It extends down to -3 dB near 15 Hz. It does not exaggerate cymbals or sibilance, a fake "hi-fi" attribute that I abhor—I downgrade headphones or loudspeakers that exhibit those anomalies. Unfortunately, a lot of headphones that get high marks have an exaggerated high end. That seems to be the trend, and I don't like it. My advice: Don't be seduced by the treble unless it sounds natural.

KatzCorner_Ep17_Graph_SystemFR

Fig. 17-2: Dynaudio plus JL subs. Left front /Right front amplitude response before correction (brown/blue). After correction (red/green). The "before" curve has been offset 5 dB for clarity.

The step response of this loudspeaker system is just as astounding (fig 17-3):

KatzCorner_Ep17_Graph_SystemStep

Fig 17-3: Dynaudio Step response, left front channel. Before correction (red). After correction (brown)

Take a look at the red curve (before correction): We notice a distinct positive-going initial pulse (which is the tweeter), followed by a negative "recovery" (the midrange driver) after which the signal returns to positive and eventually decays to zero. The initial up/down/up of the impulse takes about a ms to complete. By contrast, after correction, the step response (brown) is a single distinct, short positive pulse which smoothly decays to zero in about a ms. This is a remarkable achievement: The loudspeaker has been converted into a nearly perfect "single point" source by the magic of DSP phase and time correction.

About eight feet tall, the two loudspeakers produce a coherent planar wavefront. Bass energy moves your pants leg; you can feel it. All the partials arrive at the listener's ear at about the same time. Sonically, this translates to an impressive transient response, a very coherent image, and a big soundstage with holographic depth. The snap of a snare drum sounds very realistic. All genres of well-recorded music sound impacting, real, and natural in this room. A very pleasant experience, indeed and an excellent reference to compare against headphones. Those are my tools. In the next episode we'll put them to work.

1 For those who are not familiar with Bob's work, his discography can be found at tinyurl.com/bobkatz. Bob's recordings have been used as references at high-fi shows worldwide. Several have been judged "Records to Die For" and "Recording of the Month" in Stereophile magazine. Three of his albums have earned Grammy® awards.

COMMENTS
stanzani's picture

I would replace 'loudspeakers' by 'room + loudpeakers'

X