basic experiments

I was surprised how easy it was to create this vocel-like sound with a single resonance filter (a single transistor stage with variable feed back). Possibly the effect with the vowel change comes from the combination of the sliding notches with the pitch (my Theremin is different to the Etherwave) and the fixed resonance amplification by the external formant filter.   -- JPascal

Yes, I believe the sliding notches from heterodyning are causing the vowel changes.

"I also own the wonderful vinyl LP and the additional wave tracks."

Same here, Dorit very graciously gifted it to me when she visited a couple of months ago.

"But what would happen if the filter resonance height or frequency was controlled via the CV output for pitch or volume is another question. I suppose, the voltage was used via foot pedal and in a fix high. May be somone knows more?"

I PM'ed Ian on FB but no response so far.  It doesn't look like he uses FB much at all.

"However, I hear more of a beautiful flute and violin sound in the recording with some vocal impression in a narrow audio frequency range."

The composition is quite something!  It's too bad it didn't make it onto the vinyl version, but I suppose it was an issue of space limitations of the media.

Next step in investigating the formant filtering of the Theremin audio signal. I built an additional second variable filter with a wide frequency range on a second breadboard. By increasing the Q factor, the filters can also be used to generate tones, the signal increases by more than 20 dB, and the resonance curve becomes sharper.

Jan, I’m always very interested in your research.

How do you plan to address the issue of overload when signal harmonics fall within the formant region? This is a particular concern for the theremin, as its gliding tone means that at one frequency there may be underload, while at another there will be overload.

"How do you plan to address the issue of overload when signal harmonics fall within the formant region? This is a particular concern for the theremin, as its gliding tone means that at one frequency there may be underload, while at another there will be overload."  - ILYA

Mostly on-topic: The peak gain of a state-variable filter rises with Q (and damping = 1/Q) and if you examine the topology it's pretty clear why.  Zolzer's DAFX book shows a method to normalize this to some degree.  An input attenuator is fed sqrt(damping), and the filter damping is then just the square of this, which is trivial do in software.  I tried this method for the D-Lev formant bank, as well as for the oscillator and noise multi-mode filters, but wasn't happy with result.  Without normalization, the Q clearly interacts with the peak gain, which makes setting up the filter more problematic.  Though both controls are mostly exponential, so for every one click up for the Q, clicking the gain down twice keeps the peak level more or less even.  Having the peak gain up with Q while the skirt level remains the same just sounds more natural to my ear. 

Thank you both, Dewster and Ilya, for the feedback. 

I  will try to explain my thoughts using some experience with FFT-NMR. Broad frequency signals, which occur, for example, after very short time pulses with rapid decay, form broad humps in the spectrum that stand out from the stochastic noise. If narrow signals with a very long decay time are detected at the same time, the lines add up without influencing the broad line. The noisy hump-shaped baseline lies below the narrow signal line.

The formant filter forms a hump-shaped, broad resonance line that can be heard in isolation like a breathing sound, when white noise is the input. When you form a vowel with your mouth and breath without using your vocal cords, it sounds exactly like this. When the theremin signal is added, it lies on top of this broad signal. This does not result in overload or narrowing the filter curve itself.

Short break on the formant issue.

I would like to add something about the extreme importance of avoiding hum modulation. Hum is for example fed into the antenna rod via the hand/body-capacitance and amplified by the pitch oscillator. 

A simple LT Spice simulation can illustrate this. What you see in the image are the variable and fixed oscillators, with the variable oscillator weakly modulated at a frequency of 50 Hz. You can see in the audio signal that the fundamental and harmonics have sidebands at 50 Hz intervals, resulting in a highly distorted sound. Rooms with a lot of hum or other electrosmog are poison for the theremin. 

"I would like to add something about the extreme importance of avoiding hum modulation. Hum is for example fed into the antenna rod via the hand/body-capacitance and amplified by the pitch oscillator."  - JPascal

Indeed!  I believe it's like 50Hz / 60Hz vibrato, which would be +/- modulation of the variable frequency oscillator (i.e. FM, as is your example).  I also believe that it's there all the time due to the mains electric field being so strong in most modern spaces, though it is almost certainly increased via capacitive conduction through the hand & body. 

Outside of the extensive digital filtering in the D-Lev (4 notch filter bank, followed by an 8th order tracking low pass filter) I'm only aware of one formal attempt to minimize it, and that is in the November 1996 issue of Everyday Practical Electronics, "EPE Elysian Theremin With MIDI Box" by Jake Rothman, but it's applied to the volume field rather than the pitch field.  That article is one of the most practical and straightforward that I've read re. analog Theremin design.

The parts have arrived – prototyping of the JPascal Vowel Machine to enhance the thereminsound can now begin.