Let's Design and Build a (mostly) Digital Theremin!

Violin 2

Heartened by the relatively quick and painless improvement of the cello preset, I decided to take another stab at the violin preset.  As with the cello, I spectrally examined the lowest octave, dialed in the resonances, and the played with the Q's, levels, and finally the resonator settings.  Here's the result:

The formants are set to 285, 403, 570, 740, 1141, 1480, 1975, and 3729 Hz.  Astute readers might notice the jacking up of pitch correction strength (filt=18, pvel=9) as this is a difficult voice to control (and I'm not practicing enough lately).  See what you think: [MP3].  I think it's a significant improvement over my previous violin.  There's nothing like analyzing a real instrument, and not relying on resonance data you find in papers and books and such.

And this morning I added a "Line" knob to the "LEVELS" UI page to provide some control over the secondary audio outputs.  Had to kick the "Mute" knob off of that page to make room, but it's still on the main page.  You can see this too in the librarian pic above.

I'm starting to get a little smarter about saving the preset I'm editing at intermediate points to consecutive slots, which gives me a fall-back position, as well as something to compare the progression against.  Ears adapt and get tired and stuff, all of which works against you in this sort of endeavor - you constantly need a clean pair of ears.

I'll have to listen to and compare this new preset later, but for now I want to mention something that has been on my mind ever since I loaded your batch of presets after my "Pro" upgrade.  I hate to bring up things that suggest a software change at this late date, but your mention above of changing the pitch correction strength made me think of it. So here it is.

There are certain parameters that I had to go through every preset of the upload package and zero out almost immediately.  Those included all of the pitch correction settings and some of the envelope settings on the volume page, or basically everything that could be considered a playability enhancement that might make the instrument sound more on-pitch or in some way force altered playing technique.  Now I'll admit that I have gone back and added a little vvel and fall to the string presets, but nothing else.

Don't get me wrong, I do think having these features can be important, but I do think that it would be nice to have a global switch to turn these off without losing the carefully tuned settings that you have made part of the presets. 

I feel that this switch to globally enable or disable some of the parameters should also default to disabled, allowing all of the PITCH settings to otherwise be retained in the presets even if not actively used. And for the VOLUME envelope settings it's a little harder to make the call where to draw the line, but I would maybe have vvel, rise, and fall (and vloc and damp if they are tied to the others) globally controlled.  I can't think of any others at the moment.

I guess IMO the basic philosophy here would be to make globally switchable, and also switch OFF by default any preset parameter that introduces an envelope or pitch alteration that might startle or annoy an experienced player that has learned to control these things by skill.  It's a little tougher to decide what's right for presets such as strings, where vvel and fall can really offload some left hand effort, and these voices (which really are new to most theremin players anyway) can sound excessively synthetic without any at all.  The thing is, who decides how much is correct?  And too much is probably worse than having none, so maybe having anything turned ON (and I'm talking only about playability parameters here) that a traditional theremin would not have should instead be defaulted OFF.

I've thought about this quite a bit before mentioning it, but I don't think that having a package of presets with these enhancement functions turned on by default is a good idea.  But they should still be easily available to turn on as I would imagine that they might be on the Theremini. 

Hi Roger,

I know where you're coming from, and this sort of thing is a tough call.  My main fear would be that global disable(s) could be confusing, particularly to the noob ("Hey I'm turning the pitch correction all the way up and nothing's happening, what gives?!?"  *an hour later*  "Lordy, it's this knob over on the SYSTEM page, sheesh!"  *a week later*  "Hey I'm turning the pitch correction all the way up and nothing's happening, what gives?!?" etc.). 

Pitch correction can be killed dead (per preset) by pressing either/both the "span" or "corr" knobs to zero them out (I guess I would pick "corr" to zero out as it is at the end of the signal chain IIRC).  These knobs are normally maxed out (i.e. = 31) when in use, so that's not too hard to remember should one want the default pitch correction for that voice.

Envelope is more complex, though pressing "rise" "fall" and "damp" will kill most of it (per preset).  These would obviously be more difficult to restore to the default as they generally have intermediate values assigned to them when engaged, but seasoning them to taste and saving them that way (as you are doing) would be the most likely route users would take?

The systematic solution is to massage the presets in the editor and upload them.  Since the original presets are files, there's no danger of losing them in the editing process (if they are supplied on CD or the like).  So I guess I'm going to dodge your very practical suggestion for the time being as multiple mechanisms exist to more or less address it. 

Also, I'm imagining the the general use case will be to for the player to consult the LED tuner most or all of the time, which is rather mandatory when using pitch correction (at least when playing a cappella), and in that scenario even fairly strong pitch correction doesn't fight you too much IMO, so it would be more likely to be not disengaged, but similarly seasoned to taste and saved with the voice.  There are voices (like the accordion & clarinet) that sound fairly unnatural without strong pitch correction / quantization (unless you're a good finger-style player and can do quick intervals).

[EDIT] I'm also planning on making a more traditional GUI-based editor, which should further facilitate preset management.

Voilà, Viola!

Same approach as the cello & violin, here's my viola:

Started with the latest violin patch, dialed in the formants I saw in the viola sample lowest octave (all on the C string): 164, 213, 349, 452, 698, 830, 1397, and 1710 Hz (see pic below).  There seemed to be a notch around 1kH so I avoided placing any formants in that region.  Seriously messed with the resonator setting for at least an hour - sooo much of the overall character is due to the resonator!  See what you think: [MP3].

I'm using the spectral view in Audacity to resolve the explicit formants.  If you highlight the entire octave of notes you get humps, and varying the FFT sample size really helps to resolve them at various frequencies (you need a certain amount of samples to resolve lower frequency content, but too many samples higher up the spectrum gives too much resolution, hence the varying of the FFT sample size during this analysis):

The above is with 2048 samples, and the resonance humps and notch are relatively clear.  I really wish Audacity had a real-time FFT, like Audition did.

I will say again that these are, as always, my opinions. But we have sparred a bit on many issues since the beginning, and I would like to think that the D-Lev is better for it.  I'm not being critical, but I do above all want you to see success, because I think you and the D-Lev deserve it.  Every day that I play it I am impressed by it, and I know others (most, I would hope) would be impressed as well, and that's why I the need to offer another perspective so that you can get it out there and make the best possible impression.

Begin opinion:

If I were in your place I would be asking myself what kind of first impression I want this theremin to provide, and to whom. I think that is our fundamental difference. And I don't think you can seriously worry about a hair-pulling scenario of the first paragraph in your reply above without also worrying if the noob isn't also going to have a hard time sorting out what every other four-letter UI mnemonic stands for and what it controls.   We have at times talked about this being either a pro-level instrument or an inexpensive but high capability beginner's instrument.  I sense that you are now aiming at the latter, even though the user interface is pretty daunting for that audience.

I would argue that even a noob knows, or can quickly learn the concept of global versus preset settings.  If you don't agree with the need for global kill switches then it might make sense to segregate the presets into pitch corrected groups, as other instruments or effects processor do.  But however you do it, to provide a set of "factory" presets with everything turned on by default would be a big mistake. 

An accomplished theremin player is not going to like flipping to a preset that has pitch correction and envelope alterations.  I'm a pretty low-skill player and even my first impression with the string presets was that something was wrong.  Pitch was acting weird, and the response times reminded me of the early problems with volume response.  Then I looked into the presets and saw the envelope and pitch correction settings, and zeroed everything out. 

Pitch correction as you say only works if you look at the tuner, play with accompaniment, or are gifted with perfect pitch to provide the reference that is needed. And even then for strings and other instruments that have a lot of transitional nuance, it gets in the way more than it helps.  I don't play to the tuner, and it's not a good idea to assume that others will stare at it either during the course of playing.

I think what needs to be considered is the balance between having to tell the experienced player to turn things off or the noob to turn things on.  For a reference of how others handle this, the only other instruments that I have are two effects processors that have optional pitch correction.  It is, like chorus, delay, and several other functions switchable on a per-preset basis, and not globally as far as I know.  But a key difference with both of these is that the presets can be named and the displays show which effects blocks are active in a given preset, neither of which the D-Lev has.  And on top of that the pitch correction presets are limited and grouped.

I could agree with not having global switches, but I would have a limited group of presets with any of the "opinion-based" functions enabled, perhaps just to give a quick taste of what they do.  But I would never have any of the corrections applied to the standard instrument presets unless they are off by default.

End opinion.

"I will say again that these are, as always, my opinions. But we have sparred a bit on many issues since the beginning, and I would like to think that the D-Lev is better for it."  - pitts8rh

Same here and understood - this project has very much benefited from your input!  I don't think a better collaborator than you actually exists!

"We have at times talked about this being either a pro-level instrument or an inexpensive but high capability beginner's instrument.  I sense that you are now aiming at the latter, even though the user interface is pretty daunting for that audience."

The target has been me, then you and me, and I often think about my Thereminist friend Philip and how he might react.  But I don't really know who the target is.  The axes are hard enough to calibrate (though once the physical design stops moving this can be canned to a large degree) that I'm a little afraid of the total noob (though I they buy and use EW's, and those are a bear to tune).  Seasons don't fear the reaper, but they probably fear the noob - as should anyone who finds themselves placed in a manufacturing scenario! ;-)  Nothing on noobs, per se.

I suppose the target audience is good Thereminists who already have an instrument or three and are very familiar with how they operate?

"If you don't agree with the need for global kill switches then it might make sense to segregate the presets into pitch corrected groups, as other instruments or effects processor do."

That would be trivial, and trivial is what I do best! ;-)

"An accomplished theremin player is not going to like flipping to a preset that has pitch correction and envelope alterations.  I'm a pretty low-skill player and even my first impression with the string presets was that something was wrong.  Pitch was acting weird, and the response times reminded me of the early problems with volume response.  Then I looked into the presets and saw the envelope and pitch correction settings, and zeroed everything out."

If there were only one option, I guess turning off pitch correction for virtually all the presets (except those that rely on it as a "special effect") is the safest route to take.  I'm feeding you my own raw presets as they become available, which won't be the general case.

"I don't play to the tuner, and it's not a good idea to assume that others will stare at it either during the course of playing."

We shall (hopefully!) see!  I'm certainly not the best player, and perhaps part of that is my over-reliance on "training wheels" like the tuner, but without it I'm almost certain that I wouldn't play nearly as much as I do, and may have given up by now.  It's not just to align the notes with the pitch correction (though that's a lot of it, and a practicality I didn't anticipate at all up-front) it's also staying on-pitch through a song without accompaniment (which is my default mode, unfortunately).  Perhaps as - or even more - importantly: the tuner lets me literally see the key, where the song notes are in that key, and the intervals between notes, and this provides me with information about the structure of the song, and music in general, that I never would have obtained on a regular Theremin, or any other gliss-based instrument.  e.g. I wasn't aware how many "interesting" notes in songs are accidentals to the key, and how their placement in the melody seems designed to emphasize them.  And seeing the size of the intervals informs the size of my hand movements.  It's so powerful.  But I do get your point.

"I think what needs to be considered is the balance between having to tell the experienced player to turn things off or the noob to turn things on.  For a reference of how others handle this, the only other instruments that I have are two effects processors that have optional pitch correction.  It is, like chorus, delay, and several other functions switchable on a per-preset basis, and not globally as far as I know.  But a key difference with both of these is that the presets can be named and the displays show which effects blocks are active in a given preset, neither of which the D-Lev has.  And on top of that the pitch correction presets are limited and grouped."

Very eloquently put!  A small contiguous group of instruments with pitch correction engaged could easily be implemented.  I'd like to do this anyway because the effect can be made subtle sounding yet quite strong in effect, and it would be nice to have a few examples of such settings in there.

"But I would never have any of the corrections applied to the standard instrument presets unless they are off by default."

Excellent advice!  Thank you for taking the time to explain your observations in more depth!  I definitely see where you're coming from now.

staying on-pitch through a song without accompaniment (which is my default mode, unfortunately.

I'm almost sure you've encountered it, but just to be sure: There are a bunch of accompaniments for each element of a large subset of everything under the sun of what's considered at least somewhat well known songs, especially piano only, to sing over.
I mean played by humans who do this as a hobby, often skilfully and sometimes even tastefully played. Attempts to sing over the usual MIDI file usually make me give up in disgust.
Oh, and then there are the more "complete" ones (this channel apparently sometimes has audible singing, but at a much lower volume than the accompaniment which, I guess, is easily overpowered by one's own playing)
https://www.youtube.com/watch?v=l6aPgQjGxQU&list=PLs3ing7fJIcqSAiqXVga_DTkmMUNuUOwh&index=16

(also remember you can set a lower playback speed on a youtube vid, creates some artifacts, but not horrible at 0.75)

"(also remember you can set a lower playback speed on a youtube vid, creates some artifacts, but not horrible at 0.75)"  - tinkeringdude

Ha, I did not know that, thanks!  Often the tempo is just too fast (and I'm a crap player in need of every available crutch) so this feature is pretty handy.  I should be playing along with these things at least initially anyway, just to avoid having wrong notes and timing and such permanently entrenched in my head.

Flute

Worked for a couple of hours on a flute preset.  As usual the audio reference was from the University of Iowa anechoic samples collection.  Here's the preset:

I looked at the spectra of the lowest octave and dialed in the mushed together resonances, but there aren't mucho resonances in the flute, and so that route wasn't fruitful and I dialed them back out.  It's more of an all / odd harmonic mix thing, with some tracking high pass filtering, and some velocity envelope generation.  Here's a sample with external reverb applied: [MP3].  Flute has more harmonic content than I would have guessed.  This patch still sounds a bit too much like a recorder IMO, but I don't know if I can get it any closer with the present signal chain.

I'm running behind you so I still can't comment on the latest presets, but I've been wanting to ask a general question about how you develop instrument sounds, particularly when it comes to resonances and formants.

I understand how watching the frequency spectrum of an instrument over time (or just doing a max hold) can give a good sense of the frequency-domain envelope and from that the formant peaks can be seen.

But specific to the architecture of the D-Lev, how do you differentiate between peaks to be relegated to individual formant filters versus the resonator block?

Listening to the inharmonic resonator alone with a very low frequency impulse-like oscillator source gives the impression of a drum with adjustable tension, among other things.  This block seems to have an immediate and satisfying effect on the timbre of bright voices like strings, and a whole bank of variant string presets can be generated quite easily by simply tweaking the RESON freq and some of the other parameters.  And when I was asking earlier about tap-testing instruments such as cellos and violins I was thinking that the resonant sound of a real instrument's body would be most closely emulated by the RESON settings, even though an impulse response would show the formants as well, I think.  I'm not sure that this simplistic thinking is correct.

So does the measured or published spectrum of a real instrument have characteristic clues that indicate which peaks are best created by formants and which are best handled by the resonator block? Are you visually sorting out the fixed-frequency content that is specifically NOT harmonically related to the oscillator? Or is the RESON block set largely by ear after the formant peaks are roughed out?

I know I've looked at the impulse response of the RESON block alone, but I don't remember much about what it looked like.  I hope this isn't a stupid question, but it's early for me , and I'm not sure I could figure it out later either.