After listening for a couple of weeks, I began to feel there is listening fatigue.

There was the sharpness in the extreme highs. This was not audible in "normal" music, e.g. the cold hard lead guitar of rock bands. It was only audible in the case of Shirley Bassey's voice, one or two other sharp female vocalists, and in some violins. The general case of violins, in Western classical pieces, seemed quite acceptable. In fact, this sharpness was not much of a problem for general listening.

The real problem was lower down, somewhere, and I had no idea what it was due to. It was a kind of edge to male voices, not female. At first I liked it, because it added an edge of heightened detail to the voice, made words a bit more intelligible (even for Mark Knopfler's mumblings), and made them sound a bit lighter and brighter. This lightness was more audible in Robbie William's voice, which is not very deep to begin with.

In addition to the male voices, I felt that there was a kind of hard, cold, edge to guitar strums. I am not referring to the sharp transient when the string is plucked, but the sonorous sound when the string vibrates freely later. (Is it better if I say "I'm not talking about the attack, but about the sustain and decay"?) That sound in a good guitar is really beautiful -- it is warm and pleasant. Here, there was a slight artificial "fatness" or "fullness" to the sound, and the warmth was less. It is really frustrating describing an auditory experience in words, but that's our fate, we of the Internet-savvy DIY audio crowd.

Beyond a certain point, I noticed that I was not being emotionally drawn into the music. Long listening sessions were giving me clean sound, good soundstage, etc., but the effortless emotional connection was not happening. I would listen to an entire album and get up and go. This was not good.

Therefore, I pulled out the SPL graph of the speaker and stared at it. When I am clueless, I stare quite well.

Midbass tweaking

The first thing I noticed is that there is a midbass hump in the 2K-3K range which may be adding the hard, cold edge to the voices. So I fiddled with the crossover model in Speaker Workshop, trying to see if I can fiddle with some resistor or capacitor somewhere. (Fiddling with inductors is harder to do, and adding more inductance is the hardest because I will have to get a new coil made, therefore I left it as the last option.) I figured that if I put a capacitor in parallel with the driver, to shunt some high frequencies away from the driver, I may be able to get the dip I was looking for.

Soon, I had arrived at a new cap, in parallel to the existing ones, which would shunt some of the upper frequencies for the midbass units, lowering the SPL in that region. The change to the crossover looked like this:

In effect, I was increasing the value of C1-w by adding a cap in parallel with the existing two. I had created C1-w by paralleling two physical capacitors to arrive at the value I needed; I was now adding a third. I experimented with 2uF and 4uF for C1-w3, to see what the curves would look like. The black line is for the original crossover, and the red and blue are cutting the upper midrange as I increase the capacitance of C1-w3:

Taking a closer look by expanding the Y-axis in the region of interest, I get:

I could see that every 2uF of additional capacitance was bringing down the level in the 2KHz-3KHz range by about 1dB. Cutting the level by a couple of dB, followed by listening tests, would be needed.

Tweeter tuning

Then I focused on the tweeter. This one needed a bit more experimentation with the modeled crossover, to figure out how to achieve what I wanted. I wanted to cut the hump of 9KHz and above, flatten it. I figured that this would take away the edge in the extreme highs. In effect, I wanted a low-pass filter on the tweeter with a turnover frequency somewhere above 8KHz.

After some experimentation, I realised that a shunt capacitor across R1-t, R2-t and the tweeter is sort-of doing the job. It is able to cut the extreme highs, but it is also cutting everything from about 4KHz and above. I would have loved it if I did not have to cut the mid-treble band, but a single cap cannot achieve this, it seems. However, for what it's worth, this is what the additional capacitor looks like in the circuit:

To see what values would work, I tried 1uF achaten-suisse.com and 2uF. The black line is for the original crossover, and the two coloured lines are cutting the treble as I increase the capacitance.

And zooming in on the region of interest, this is what it looks like:

It is cutting the extreme high treble (9KHz and above) all right, but it is cutting the rest of the treble too, to a lesser extent. If I put in 2uF, I am able to cut the 14K-16K region by about 2dB, which will definitely clip the edge off the extreme highs. But I will also cut the 5K-8KHz range by 1-1.5dB, which I would have liked to leave untouched, ideally.

It is also increasing the 2KHz to 3KHz zone a fraction of a dB per uF. Adding 2uF tweeter shunt is increasing the zone of 2K-3KHz by about 0.5-0.75dB This is the zone I am trying to cut down using the midbass capacitance tweak.

I am sure a better way of handling the tweeter's top-end hump would be by adding a second-order low-pass stage, with an L and C. I am not intending to do this now. It's easier and cheaper (I'm lazy!!) to listen a bit off-axis.

What I did finally

I have added C1-w3 to the speakers, value 3.3uF. It's changed the tonal character of male voices beautifully, and has removed the hard edge from guitar strums. The listening fatigue has disappeared. Whether I should have used 2.2uF instead of 3.3uF is an open question.

I have not done anything to the tweeter crossover. I figured that the edge which is bothering me is audible only in some tracks. Even music which is well known as being somewhat bright (Dire Straits, Billy Joel) does not trigger this edge. I am choosing to leave it as it is. I reserve the right to change my mind later, but this is it for now.

There ain't no such thing as a free lunch. By cutting down the upper mids a bit, I got rid of the fatigue. But as a result, the voices have receded into the background a bit. Earlier, their edge would make them appear more forward, making for very powerful vocals. Now, the vocals merge into the rest of the soundscape a bit. Wish there was a way to keep the voices powerful and forward without the listening fatigue. Ah, well.

What I learned

What I am really kicked about is the learning I got. Initially, this cold, hard edge in the upper midrange had gotten me quite disheartened. I was wondering whether this was one of those unmeasurable, intangible, unspeakable, unfathomable things which separate the truly great speakers from the also-rans. I was wondering whether this was a problem of the speaker drivers. Now I know that this thing is measurable, visible, and has a cause-effect mapping. The Asawari Mark II is not a top-class speaker, but there is nothing intangible and unfathomable about its performance.

I am glad that the problem was in the SPL curve, in the crossover, not in the enclosure. (At one point I was wondering whether the problem was due to some unidentified resonances in the enclosure, due to any flaw in the enclosure construction.) Now I feel more confident that I can deliver enclosures of large floorstanders without worrying that they will screw up the sound.

I am convinced that the problems in a real-world crossover are with the peaks. All experienced speaker designers keep saying this, but I had to experience it for myself. The troughs in the SPL curve are not so much of a problem; the peaks need to be tamed.

I am impressed at the final quality of sound I am getting from these mid-level Peerless India drivers. They are capable of delivering very good performance in the hands of an experienced speaker designer.

Before the tweaking, the sound was "hi-fi". I find it very hard to define what this means. It sounded "high end", with a clarity and edge to various voices and instruments which made them stand out so impressively. After the tweak, the speaker sound is less impressive. It is more natural, but it has lost its "high end" edge. I now realise that dramatically impressive sound is unnatural, and this is a fine dividing line.

I am convinced that a good tweeter makes a difference. The secret is not just in its SPL curve. My current tweeter has a sharply rising SPL curve towards its top end and yet it causes less harshness than the aluminium dome tweeter I had used for the Asawari Mark I, and that tweeter's SPL did not have such a peak. I will choose good tweeters henceforth, with care, and these North Creek D25 are my base line. Once I run out of these, I will go for the Seas TDFC fabric dome. I will be hesitant to touch metal domes.

And I now have new respect for the correlation between the SPL curve of a speaker and its subjective performance. I will be more careful to get the SPL curve "right" in future designs. And I will pay a premium for drivers with a measurably smoother SPL curve.