Electronic and Cast Bells



I was intrigued to hear some of the bell sounds offered by "electronic carillon" manufacturers, and thought it would be interesting to investigate why they sound so different to real bells.  Easy enough to do in principle - download the sound of an e-bell and compare it to the sound of a real one.  Nothing's ever that straightforward of course, but even the difficulties one runs into are worth recounting.

The e-bell

The electronic carillon I chose to examine is one manufactured and marketed by Schulmerich Bells, of Sellersville, PA, USA.  And the particular voice is titled English Bells.  You can hear the sample I used on their website, rendering the beautiful Coventry Carol (Lullay, lullay, Thou little tiny child), a piece so wondrous as to deserve its own Wikipedia entry.  Here it is played by Schulmerich's English Bells.

Extracting a note to test rapidly becomes a choice of one.  Only the first note is suitable, as every other note contains remaining vestiges of the notes before it.  Bb it is.  And since the second note kicks in at 0.7 seconds later, we have just 700mSec of sound to play with!

The real bell

Since the Schulmerich e-bell was described as an English bell, it's logical to compare it to a real bell from that country.  Conveniently, I had a recording I'd made of the bells at the National Carillon, Canberra Australia.  The bells there are by Taylor Bells in Loughborough, England, so fit the bill perfectly.  The wind was high on the day I recorded the bells, so there is some background noise.  But at least I have the full bell strike and decay to play with.

Compare the sounds

I've edited together this comparison of the sounds.  Firstly, you'll hear three strikes on the Taylor's bell at Canberra, followed by its decay, and then three strikes from the Schlumerich e-bell.  I had to fiddle with the end of the recording.  Because I only had 0.7 seconds of useful sample, the recording would have stopped dramatically but disconcertingly after the third strike.  I did the only decent thing I could think of - add a fourth strike and fade it rapidly, to disguise the abrupt termination.



Quite a dramatic difference in tonal qualities, I think you'll agree.  That should mean our analytical processes should also have little difficulty in identifying the obvious differences. 

Unfortunately, one of my favourite analytical tools, WaveAnal's Partials Decay section, refused to have anything to do with the Schulmerich sounds, probably because it quite reasonably regarded 700mSec as not enough to work with. 

It's a shame I didn't have access to the full decay of the e-bell as I suspect Decay Analysis would have an interesting story to tell there too.  The partials of real bells attack and decay at different points in the strike, so the tone develops as it progresses.  I'm not hearing that in the Schulmerich emulation, but, without a longer decay, it would be difficult to confirm analytically.

WaveAnal's automatic Partial Identification system was also struggling, for reasons we will see in a moment.  But it was able to process and export the FFT data in a form I could deal with, leaving me to struggle with the identifications! 

I ran both sounds through the system and used Excel to prepare the comparison below:

As you can see, the results are very different!

The real bell, Canberra Bb, is shown in pink.  (I've dashed it so as not to conceal any overlaps with the e-bell in blue.)  You can see only four partials, the Hum at just under 1000Hz, the Prime the small blip just under 2000Hz, the Tierce at about 2200 and the Nominal at about 3750.

The e-bell is represented by 7 partials.  You'll see only one of them is at the same frequency as the real bell's, and that is the Hum just under 1000Hz. There are also two partials below the Hum (decidedly abnormal!), and the other partials are at very high levels compared to the real bell's.

In musical terms, please!

Enough of this geeky Hertz stuff, and tell us in musical terms, I hear you grumble.  This table might help...

Partial Name Expected Canberra Schulmerich Schulmerich Deviation
      Db +26 +26
      Db +14 +14
Hum Bb Bb -2 Bb +2 +2
      F +7 +7
Prime, fundamental Bb Bb -3 B -6 +94
Tierce, minor third Db Db +5    
Quint, fifth F   E -6 +94
Nominal, octave Bb Bb -1 Ab +28 -172

The first two columns give the names of the usual partials, and the pitch we would expect them to be at.

The third column gives the pitches of the four partials we detected from the Canberra bell sound, and their deviations.  All as expected and within a remarkable 5 cents.  Note the absence of the Quint.  It may well become more prominent at Forte playing levels.

The fourth column gives the Schulmerich e-bell results.  As we saw in the graph above, only the Hum partial seems to be where we would expect.  But note that the other pitches we expect (Db and F in the case of a Bb bell) are represented, although not where we expect.  The Db's are below the Hum (unheard of?), and the F is just above it, rather than in its expected position an octave higher. 

Legitimate in a certain sense.  If a bell note is really a chord, then this is an inverted chord.

But wait, there's more!

But we have a bit more to explain.  Schulmerich's nearest candidate for a Prime is a B, not a Bb; its Quint is an E, not an F, and its Nominal an Ab, not a Bb.  To make that all a bit clearer, I've added a fifth column which computes the deviation (in cents) of each partial from where we might have expected it.   Remembering that 100 cents is a semitone, some of those deviations are substantial.

So why?

My guess is that the e-bell sound is synthesised, and is not a real bell sampled.  Once you decide to synthesise, you are freed from the constraints of the real world, and can put your partials anywhere you like, and make them the size you feel works best.  So this is a construct arrived at in cold blood. 

I imagine the decision to lower the pitch of the Db and F partials was in the aim of giving the bell greater gravitas.  I can't put forward a reason for the large deviations, excepting to remind us all that bell pitch is largely a psycho-acoustic construct, not a simple matter of physics, and it might be that extensive listening tests confirmed these parameters as the most pleasing.

Capable or Culpable?

Carillonists are commonly scathing about electronic carillons - their endearing descriptor "bongatron" is enough to confirm that.  Whether the analysis and comparison above adds fuel to that derision is in the eye (and ear) of the derider.  I imagine Schulmerich went to a lot of trouble in designing that sound, and, as presumably it sells, they must regard it as successful. They could probably defuse some of the discontent if they did not attempt to pass the sound off as something it quite clearly isn't - the sound of an English Carillon. It isn't even the sound of a bell.

I should add that the choice of company and sample used in this article is by chance only.  Further, that this comparison cannot be extended to other voices by that company or other companies. 

Back to McGee-flutes Index page...

Created 18 April, 2013