OMG you guys! All of these talks about the TM ~ making me want to get my hands on it! I wish I could even think about this, but unfortunately now, all I think about it to keep on playing to get better at this for a long while and then, maybe try something like this... I'll keep reading and watching in the sidelines and witness what you guys are coming up with :)
I will never learn to play the theremin at any level of acceptance but the TM is another gift shared by a master that I can give away. The theremin has a potential only a few will ever fully realize.
Before this week I mentioned the enlightened soul of the theremin and this week we all got to share in something special. In my ten years of theremin research this kind of moment is rare.
Today is a special day..........
“If you play in the soprano range the Talking Machine will not sound like a tenor. The only formant that I have found useful for the kind of thing I do is the open "AH". The others are all a little bit clownish. “ – Peter
If one looks at the formant chart :
You can see which formants are likely to work best for particular pitches – the most important being the first formant (X on this chart) and one sees that ‘ae’ and ‘ah’ (and a bit of ‘uh’) extend above 1khz.. ‘oo’ is the formant with the lowest frequency components (both 1st and 2nd formants) then there are formants (‘ee’ for example) where the first frequency is low (<= 500Hz) but there is a large difference between 1st and 2nd formant frequencies (for ‘ee’ 1st frequency is < 500Hz, 2nd frequency > 2kHz) and one can get soprano on these if using only 2nd, 3rd … formant frequencies (the listeners brain ‘creates’ the missing 1st frequency, I believe)
The formant chart shows the primary reason for this – effectively the formant filters are bandpass filters – the highest frequency passed through will be the frequency of the highest formant ‘synthesized’ – With a simple 2 formant filter, you wont get much above 3.5kHz.
I think (from the quality I am hearing) the TM must be synthesizing at least 3 formant frequencies, possibly more.. The 'muffled' "old radio" quality will be directly (inversely) related to the number of frequencies being synthesised for each formant - The TM (to my ears) has enough high frequency that it does not sound 'lo fi' at all - I am sure there are at least 3 formant frequencies being generated - But I suspect there are probably 5.. When I get back to my lab I will run a sample through a spectrum analyser - it should be quite easy to see on a sustained note, there should be a group of peaks one can easily count.
"When I get back to my lab I will run a sample through a spectrum analyser..."
Looking at the entire "goldie&tm.mp3" file in Adobe Audition (Frequency Analysis, linear view, Blackman-Harris window, 65k FFT) I see the following peaks:
60Hz @ -50dB
325Hz @ -24dB
662Hz @ -24dB
1kHz @ -35dB
1.3kHz @ -43dB
1.6kHz @ -47dB
1.9kHz @ -44dB
2.3kHz @ -36dB
2.6kHz @ -54dB
If you squint there are broad "peaks of peaks" around 500Hz and 2.3kHz.
The real test would be to feed a few seconds of white noise through this baby and post an MP3 of that. Peter (or any other owner of the TM here) could you perhaps oblige us? If so, I would be happy to post an analysis picture or two of it.
Fred said: The formant chart shows the primary reason for this – effectively the formant filters are bandpass filters – the highest frequency passed through will be the frequency of the highest formant ‘synthesized’ – With a simple 2 formant filter, you wont get much above 3.5kHz.
Yes, I think this is the problem. I have just made the test dewster mentioned, but for the Guitar Rig filter instead of for the TM (which I don't have... yet). I first generated some white noise with Audacity and obtained the following spectrum analysis:
Then, I opened Guitar Rig, passed the wav file with the white noise through the formant filters with the settings that I had used in recordings 2, etc. yesterday, and back in Audacity I got the following analysis:
So it seems that it's boosting three frequencies:
588 Hz @ -26.3 dB
1127 Hz @ -19.2 dB
2488 Hz @ -19.1 dB.
Looks like it's doing an "ae" in the table you posted, and adding an extra frequency... but as you say, nothing above 3.5 KHz (well, a downward slope, but I guess that doesn't count, right?)
I'm learning a lot in this thread.
"The real test would be to feed a few seconds of white noise through this baby and post an MP3 of that." - dewster
The above would work (probably) only if the TM 'simply' processes the input through a set of bandpass filters corresponding to the selected vowel - But I suspect there is more going on than that - I suspect that some analysis is being performed on the input signal, and that the filters are being adjusted to best match the input signal.. Also, I suspect that additional harmonics are probably being synthesised from the input signal if this signal lacks the minimum required harmonics with which to create a vocal sound.
Each vowel has quite a wide range of possible frequencies for 1st and 2nd formant - take 'ah' for example, 1st formant could be between 700Hz and 1100Hz and 2nd between 100Hz and 1600Hz, and it would produce (for these different frequencies) variations of 'ah'...
I suspect that the different tonal quality which results from input of different frequencies and waveforms may be more complex than simply being the result of fixed frequency bandpass filters 'responding' to the signal -
This has been one of the challenges I faced with analogue formant filters - It is quite easy to get good quality voice effects when one has a constant input frequency with optimum waveshape, the problems start when one applies varying frequencies to make it sing rather than talk - To get enough harmonics into the filters with a varying frequency, one needs an extremely 'harsh' waveform - And unpleasant harmonics tend to break through the filter when one does this - the solution (increasing the filters Q) gives an artificial quality to the tone... The 'solution' I went for was to have a less 'rich' excitation signal which eliminated unpleasant harmonics, but also weakened the formant - it became more of a tone 'coloration' than a true vocal simulation.
This is why I think something more is happening with the TM - IME, Even a 'rough' theremin signal does not contain enough harmonics to generate the formant clarity one hears from the TM. Listening to the TM with distorted guitar input, I can hear harmonic breakthrough, but do not know what the wet/dry mix is so cannot be sure about what I am hearing.... The best vocal synthesis I have heard from the TM has certainly been what Peter has done.
Feeding noise into the TM will certainly be interesting - But I think the results will depend on how noise is interpreted by the pre-processor.. It may be that the formant frequencies will be quite different to what they will be for an input signal with a fundamental frequency.. The real test would be to feed ramp, triangle, square and sine waves of differring frequencies into the TM.. Sine would be particularly interesting, because any harmonics would need to be created by the TM.
"but as you say, nothing above 3.5kHz (well, a downward slope, but I guess that doesn't count, right? )"
Well, the 'downward slope' does count - one can see that there is no emphasis of any frequency above 3k5 (last peak at about 2k5) but there is certainly audible frequency above 3k5.. The signal is being progressively attenuated as the frequency increases, as is normal - and this depends on the slope of the filter.
Noise has constant amplitude over the spectrum - This is not the way other signals behave.. But this plot demonstrates the problem one can have with harmonically rich waveforms - the upper harmonics can be heard because the filter cut-off is not sharp enough.. And if these harmonics are unpleasant and strong, they can spoil the ointment.. Its fine if the harmonics are musical and subdued.
I suspect that as well as having its resonant peaks, the human voicing mechanism also has its natural 'roll off' and that larynx produced signals will naturally tail off, but still be audible well above 3k5Hz - Well, I know this must be true - vocal microphones need to be flat to about 9k to be any good..
So to perfectly synthesise voice, one needs to perfectly synthesise the larynx waveform at all frequencies, and perfectly synthesise the vocal resonances at all frequencies - and to do this DSP is needed, and it may be that the TM is in fact doing physical modeling in a more sophisticated manner than simply employing formant filtering.
Borrowed a MOOG Etherwave Theremin (not the pro, unfortunately) from a friend and am starting to learn how to play. I sound like a really remedial violin student! But I am hoping since I sing, it will come. It is just finding the right place for the note to be right and in tune. Got to watch more Youtube! The sound on this thing is dreck, but at least I can start to develop muscle memory and coordination.
What is a good stand that is sturdy for a theremin? The one I have is a weighted mic stand and it lists. Are there any good weighted ones? I don't like the tripod one because I trip over it.
I have never had any issues with my Proel OST110CR. (One point worth noting is that I have it on its lowest setting and I'm six feet tall, so it's not recommended if you're the last one to get wet when it rains.)
I also have a generic stand with a slightly smaller, slightly lighter base and a rubber ring and I've never had any problems with that either. I am a bit puzzled why your loaner would list - etherwaves are fairly well balanced (unlike their players) so that shouldn't happen. Erm, is your floor level?
I have my subscope on a mic stand with the circular heavy base. Works wonderfully as it is a bit heavier and not in the way when I need to move around... Alson not drinking wine while playing is always a good thing ;)