Dr. Earl Russell Geddes

Sound quality has become and will continue to grow as the single most important design criteria for any audio component.  It is important to understand this concept and what it means to our products.

Backround

I will first give some background on my own involvement in Sound Quality and how I came to understand it's importance and what I think its role is in the marketplace.  This section will be mostly historical.

In about 1981 I was introduced to a gentleman who was, as it turned out, the Chief Technical Officer for Daimler-Benz Cars (Mercedes) in Stuttgart, Germany.  While that was not his actual title that is what he would be called today.  For several years, he and I met at the annual Society of Automotive Engineers (SAE) convention in Detroit.  He was particularly impressed with some papers that I had done on automotive sound systems.  After several years he invited me to visit the Research and Sound Labs at Daimler-Benz, which I did in 1984.

What I saw at those labs impressed me very much, they were taking a completely different approach to sound (noise control) than I had ever seen before.  Basically, they were not concerned with the level of the noise - in a Mercedes it was already low - but they were very concerned with what they called the "quality" of the sound.  This was an entirely new concept to me - Sound Quality.

On my return I wrote a trip report highlighting what I had seen.  This report was very widely circulated at Ford Motor Company, it went to the very top of engineering in a very large company.  In this report I highlighted my impressions about "Sound Quality" (I will us SQ from now on), and how SQ was a fundamentally different approach to noise control than what was being used at Ford, or anywhere else in the US, at the time.  The eventual outcome of this was that I was asked to create a new "process" for noise control at Ford, one that encompassed SQ.  In 1991, after SQ had not only become widely accepted at Ford, but at all of the major automotive companies, I was awarded the Henry Ford Technological Achievement Award - the highest honor bestowed on an engineer - for this work.  I left Ford in 1993 after achieving the highest ranking "specialist" in any area of engineering.  That is when I went to Knowles as Director of Research.

At Knowles I continued to be interested in SQ, but now it was being applied to hearing aids.  Having met Prof. Lidia Lee (who later became my wife), we did several internal studies at Knowles and a few papers together on the perception of noise in hearing aids, which is basically a SQ issue since it involved perception.

I left Knowles in 1998 and went back to Ford Motor, where I headed up the Audio Research group.  One of my main areas of interest was SQ, but now in its application to Audio products.  My Career at Ford and later Visteon was short because of the inherent trouble within the automotive world and the inevitable bankruptcy of Visteon some years later (they are still in bankruptcy).

As a consultant from 2001 to the present, Lidia and I have continued doing fundamental research into the area of SQ.  SQ has become so widely important in audio that it is now the sole topic of a special AES conference in 2010 in Sweden - engineers worldwide are discovering the relevance of this topic to their designs.

What is Sound Quality?

I think that this is a particularly important question to ask, especially in China.  The reason that I say this stems from a talk that I gave on SQ at the Chinese ALMA Symposium in Shenzhen in Oct. 2007.  By talking with several of the participants after this talk I learned that the Chinese participants really had no idea what I was talking about.  They were completely unaware of the concept of SQ nor what relevance it might have to them.  Several of the western attendees were quite interested however.  I thought at the time that I might be able to interest some Chinese transducer manufacturer in hiring me to help them develop SQ at their companies, but none were interested.  As I said, they just did not see the importance of what I was doing or what it meant.

It is critical to understand what SQ is and why it is important and a few examples are in order.

In 2001 Lidia and I wrote a book entitled Audio Transducers.  In that book, there were chapters on psychoacoustics and distortion. (When I say distortion in this paper I mean non-linear distortion, the generation of spectral content in the output that is not present at the input.  Audio systems can also have linear distortion, which means that the frequency weighting of the output is different than the input - linear distortion is also called frequency response.) In our book Lidia and I hypothesized that there would be a complex relationship between the ears ability to detect distortion and the way that it is measured.  After the book was completed Lidia and I set-up a test to quantify and validate this hypothesis.  This test was done in 2003 and published that same year.

In this test we manipulated a musical passage in several different ways and presented them to about 30 subjects in a random fashion and asked them to rate the stimuli. The distorted samples were directly compared to the undistorted sample and the subjects rated the audibility of the distortion in this comparison.  The results of this test were quite profound and continue to shock the audio world.  They have been duplicated and confirmed several times since by numerous researchers, hence validation of our results has been completely confirmed.

Basically (the complete test is available on my website www.gedlee.com/distortion_perception.htm) the results showed that there was no correlation between any current measure of distortion - a metric - and how it is perceived by subjects.  Total Harmonic Distortion and Inter-Modulation Distortion (THD and IMD - the traditional measurements) were used as a measure as well as a new derived measure which we called the GedLee metric.  THD and IMD were found to not be statistically significant, although they were weakly negatively correlated with perception.  This means that people actually like THD if they can hear it at all, however, the reality is not actually this simple.

The GedLee metric was shown to be highly correlated with the subjective perception and statistically significant with a 99% confidence.

The first thing that should be observed from the previous several paragraphs is the use of several terms and concepts, like correlation, statistically significant and confidence, that are not typically used in audio discussions.  These terms highlight the different point of view found in SQ.  SQ is based in subjective perception, which is not and never can be an absolute thing.  The basic tools of SQ are psychoacoustics and the foundations for its principles are determined through psychoacoustic research.  This type of research uses subjects, usually in large numbers, and highly controlled psychological based testing methods.  The results in these types of studies are never an absolute confirmation of anything, because basically, no two people are exactly alike and only wide generalization are possible.  This is typical when human perception is involved.

Psychoacoustics is phrased and studied in a different context than physics.  It uses correlation and statistical analysis to determine not what any one individual believes, but what the most likely belief of a random individual, or a large group is going to be.  No two people ever perceive things in exactly the same way, but in many areas it is found that the "majority", or "most", people perceive things in very similar ways.   What is meant by "majority" or "most" becomes more quantified when we look at concepts like correlation and statistical significance.  Therefore, the first step in the road to SQ understanding, is to master this new language, how it is applied and most importantly how it is developed.

It is important to reiterate that SQ does not have a meaning on an individual level.  SQ is what we expect the "average" or "typical" person to perceive, even if no such real person exists.  Thus, SQ may or may not be what any particular individual perceives and no one persons impressions can ever be SQ.  This means that there are no "golden ears" in SQ.  It is precisely the opposite concept - the perception of the group and not the individual.

Returning now to our distortion perception study, there are several key things that have come out of this study.  First, it is entirely pointless to do THD measurements and even more futile to do designs to lower THD.  Does this mean that we can and should make highly distorted products?  The answer is "Yes and No".  It all depends on the kind of distortion, not on the THD measure.  This idea of the kind of distortion is entirely new.  Quantitatively, kind can be equated to the nonlinear order - an obscure concept that has never really been used in audio before but is the only valid way to look at nonlinear distortion. 

For example, in a transducer the design of the displacement response is often made to be symmetric about the rest position to "lower THD".  But is this the best approach?  Given our new understanding of the perceived situation as described above (THD is meaningless), it is more than likely not the best approach.  This "low order" type of nonlinearity is virtually inaudible and this knowledge could be used to substantial benefit in a modified design.  On the other hand, there are types of nonlinearity that can be quite audible, like a sharp drop-off in cone motion as the coil leaves the gap for example, or a small particle in the gap.  These can be quite objectionable, and can often have very low THD numbers.

Moving on to another example, one that Lidia and I did for B&C Speakers in Florence, Italy,  we performed a SQ study of compression drivers, where we compared several manufacturers drivers for frequency response and distortion.  We found that the subjects could not detect the distortion in any of the drivers at any sound level up to the thermal ability of the units, but the subjects were able to detect the frequency response differences.  This meant that new designs need to pay close attention to the frequency response but need not worry about nonlinear distortion.

It is widely held, and true, that compression drivers do begin to sound bad as the output level gets higher, but, somehow, the study above completely contradicts this.  This led Lidia and I to do a further study on the audibility of certain types of delayed signals and short time reflections as can occur with diffraction effects from cabinets, etc.  We found that subjects were sensitive to these types of aberrations, but most importantly, that the sensitivity was highly dependent on the playback level.  This means that the ear itself is nonlinear and that much of what we think is "distortion" in the audio system is very likely to be "distortion" in our hearing.  This is a profound idea that changes many of the things that we thought we knew about "good sound".

Another example has to do with what we usually call the axial frequency response.  This is probably the single most common measurement done on a loudspeaker.  But studies have shown that the room itself has a very strong, almost dominate effect on the perception of a loudspeaker, and how the loudspeaker interacts with the room is a key aspect of how it sounds.  The reflection pattern of the sound at the listening position, etc. are all very important, but the axial frequency response tells us nothing about this situation and hence it tells us almost nothing about how the loudspeaker is going to be perceived in a real room.  One needs to look carefully at the complete sound radiation picture, not just a single axis.

What we see here is a very perplexing situation since it turns out that the two most common ways of assessing a loudspeaker, the axial frequency response and its nonlinear distortion (THD), have very little to do with how that device will actually be perceived by a listener.  In other words we have been doing and looking at all the wrong things for a very long time.

How does any of this affect things in the real world of products?  When used properly this new information can actually have quite profound effects. 

Over the last eight years or so, as I began to better understand SQ in loudspeakers, I put much of what I was learning into effect in the designs of my own line of loudspeakers.  For example, I understood that the frequency response that we perceive is far more than just the response "on-axis", but it is, in fact, far more heavily weighted on the total radiated response, i.e. the polar response.  My designs have a very high degree of control over the polar response (using tools and techniques that are not readily available or know) but I pay very little attention to the axial response.  I virtually ignore distortion (THD and the like), but have found that "thermal" effects in the voice coil are quite audible (but no one else even looks at this aspect).  From Lidia and my studies we have concluded that diffraction from cabinet edges and horn edges can be very audible and these have been minimized in my designs.  None of these new design features are difficult or expensive to do, it's just that no one else seems to know how important that they are, and so no one else does them or looks at them.  In other words, my understanding of SQ has led me to do things that are quite different from what traditional design approaches tell us.  My speakers are unique, difficult to copy, and many aspects have been patented.

Has my Sound Quality approach to design made a difference?  It has made a major difference.  There are a lot of companies that claim that they have "the best loudspeaker in the world", but my designs really are.  Looking at the reviews of my speakers (www.gedlee.com/downloads/recommendations.pdf) it is apparent that the SQ is very good.  But what has to be understood is that these posted reviews are all the reviews, there were not any bad ones, which is quite uncommon in audio.  It is very rare to find virtual 100% agreement that a certain speaker "sounds great" or "the best", but yet that is what my designs have done.  Virtually no one who has heard them has bought anything else.

My sales are not huge, but the growth in sales and customer loyalty is extremely high.  It's a difficult thing to break into a market as heavily controlled by the existing brands as audio, but very clearly I was able to do that, because I applied science to the process.  This is something that is actually very uncommon in audio because it tends to be dominated by outrageous marketing claims and what I like to call "Voo-Doo" - the blind belief in something which has not and cannot be proven or disproven.

There is nothing in audio that cannot be quantified and optimized through engineering - there is no "magic" in audio.  But the marketing approaches of the past have been so successful that there has been no reason to apply science to the situation.  However people are now starting to realize that they cannot believe the marketing claims (hi-end audio sales are way down), they understand that truth is almost non-existent in these audio marketing claims.  Hence, they, of course, disbelieve my claims as a result.  The thing is that my claims are actually true, they are not made up, my products really are better.  People who audition my speakers are always quite surprised because "they are better".  Those who have auditioned my speakers are almost shock - "they really are better".  They almost cannot believe it as they have come to expect nothing when told to expect something.  When the general audio consumers learn about this, as they are now, then everything will go my way, because they will no longer believe my competition, but they will believe me.  I told them the truth and gave them a better product because I did the engineering to really make it better.  My competitors just gave the same old product, but spent a lot of money telling the customer that it was better, but in fact it wasn't.  This has worked for awhile, but it won't work for much longer.

This is the key point of this entire paper.  Clearly, I believe that New Jialian must begin to develop Sound Quality as a fundamental core competency within the company.  In the products that you sell SQ has not yet made a major impact, but it will, just as it has in the home audio world, car audio and automotive noise control.  Once again, consider the fact that the AES is holding a special conference on this topic, in Scandinavia (www.aes.org/events/38/), and headed by researchers at Nokia and the like.  For the kinds of products that I see being made at New Jialian the SQ design aspects are not known.  Some may be known but many more are clearly not understood.  The leader in SQ in the small speaker business is going to have to do the fundamental development work to discover the SQ laws that govern the perception of these devices.

Sounds simple - but it is not.  One reason that sound quality has lagged so far behind in audio is the tremendous difficulty in doing sound quality studies.  They are long and costly to produce and require unique facilities.  Results in terms of design can be years away, years of high expense with little immediate payback.  But the company that gets to the end of this road with the SQ results to support new product lines, will win the customers, and usually at a premium price.  Further, it will be very long and costly for the competition to catch up - this makes it very easy to stay in the lead.

New Jialian is a follower at the present, but with a strong commitment to SQ research it can become a leader, perhaps the leader in small transducer technology.