Production Notes: Why My Singing Bowl Sounded Thin — The Overtones Were Playing Twice

I wanted to recreate the sound of a singing bowl.

But no matter how I built it, it came out thin.

That thickness a real bowl has, the way the tone wavers as it stretches out — I could not get there. I was stacking clean integer harmonics, and the result was thinner, not fuller.

It took a long time to find out why.

The culprit was my own code

When adding harmonics, I was using two layers.

• the base tone layer
• an extra overtone layer for “bowl character”

And both layers were generating the same harmonic frequencies.

The 1st, 2nd, 3rd and 4th harmonics were fully duplicated.

What happens when the same frequency plays twice with a slight phase offset?
It cancels.

Measured, the worst case was -9.7dB.
The overtones I had added “for thickness” were the thing removing it.

Worse still: the cancellation gets deeper the closer the two layers are in level.

The more carefully I balanced them, the more disappeared. No wonder it thinned out whenever I tried harder.

A second hole, in the reverb

While chasing that, another problem surfaced.

I had set the reverb depth to -16,
but the actual wet signal was landing at +13 to +16dB.

Specified and actual were nearly 30dB apart. I had been shipping tracks where the reverb was louder than the dry sound.

The cause: the reverb generator normalised by looking only at the single loudest sample. The total energy of the tail was left unchecked.

At first I assumed it affected one setting. When I checked, all of them were broken.

On a phone, even more was being removed

The third one was the hardest to notice.

To make the stereo tail, I built one channel as a delayed copy of the other. Through headphones this gives width and sounds fine.

But on a phone speaker, in mono, left and right are summed. Summing a signal with a delayed copy of itself creates comb-shaped interference that removes specific frequencies.

Measured, the 528Hz in one of my tracks was down -8.9dB.

You cannot catch this while checking on earphones. It only appears through a phone speaker.

I did not rebuild the published tracks

Honestly: by the time I found these, tracks were already public. Several in the Four Gods series went out with a different wet level than intended.

I chose not to rebuild them.

Two reasons.

First, they were not broken to the ear. The numbers were off, but the result held together as music.

Second, fixing them would make them different tracks. Replacing a sound people have already been listening to felt wrong.

So I recorded what happened and fixed it going forward instead.

What is still not fixed

Not everything is solved.

There is still a phenomenon where narrow frequency bands in the tail rise or drop unpredictably. Play a pure tone and, depending on the settings that day, it swings between -11.9dB and +7.1dB.

I tried smoothing it out — and the reverb time stretched from 1.10 to 2.52 seconds, turning it into an entirely different space. That attempt failed.

For now I avoid it by not pushing the reverb deep. It is not a real fix. I do it knowing that.

What I learned about measuring

Chasing this taught me two things about measurement.

Never measure with a single setting. Measure once with one arbitrary seed and a genuine 3dB slope disappears under ±10dB of noise.

Summing to mono can invert the conclusion. More than once, a mono measurement pointed the opposite way from reality.

When your ears say “something is off,” they are usually right.
But you cannot tell what is off without measuring.

Making sound turned out to be far more mundane than I expected.

Far more time goes into working out why a sound refuses to behave than into thinking up beautiful ones.

Still — when it is fixed, the sound genuinely gets better.