Half a croissant, on a plate, with a sign in front of it saying '50c'
h a l f b a k e r y

idea: add, search, annotate, link, view, overview, recent, by name, random

meta: news, help, about, links, report a problem

account: browse anonymously, or get an account and write.



Largely analogue electromechanical voice calculator

Speak to it and it tells you the answer
  [vote for,

Something i need to get out of my head. I was hoping it would come out in a more coherent and organised manner but frankly, i simply lack the knowledge and ability to flesh this out any more than i have, so as usual i’m just going to dump it here.

This is a partly analogue electromechanical device for performing simple calculations. To the user, it appears to consist of an intercom with a button, a box and a speaker. The user operates it by pressing the button, speaking a digit, releasing the button and repeating until the end of the first number is reached, pressing another button to enter the number, repeating the process for the second number, then speaking the operation to be performed and entering that in the same way: “add”, “subtract”, “multiply”. The digits of each number are spoken from the least to the most significant and are assumed to be positive.

This next bit is somewhat like the IBM Shoebox, and my extreme ignorance of electronics is probably about to show. In fact, right now i am quite shockingly unable to remember anything, so this is not going to be good.

As each number is spoken, three analogue filters analyse the sound in terms of whether it includes periods of near-silence, white noise or relatively pure low, middle or high tones. The exact pattern also depends on accent. Some words have one syllable, others two, giving them a different shape - a single sound which is relatively loud or two sounds separated by a relatively quiet interval.

It seems to me that the different features which need to be detected would be: High pitch on a relatively pure tone, low pitch on a pure tone, maintained pitch on a pure tone, noise rather than purity, a period of near-silence and a period of relatively low sound intensity between two periods of relatively high sound intensity. I also suspect that these features can be detected by analogue means.

The start and end of an utterance do not need to be detected because they are simply achieved by pressing the button on the intercom, that is, turning on the machine which detects the sound of the digit and turning it off. This would probably need careful timing.

My accent is non-rhotic, near-RP British English most of the time and this assumes a similar style of speech, though similar things could be done with other accents.

It seems that the digits differ as follows, in my accent at least:

Zero: High pitch, decreasing intensity (i.e. two syllables), low pitch. One: Low pitch, higher pitch, low pitch. Two: White noise, low pitch. Three: White noise, high pitch. Four: White noise, low pitch. Five: White noise, rising pitch. Six: White noise, high pitch, silence, white noise. Seven: White noise, high pitch, decreasing intensity, high pitch. Eight: High pitch, silence, white noise. Nine: Low pitch, rising then falling.

Except for “two” and “four”, each of these is different and a series of capacitors of different capacitance would lead to appropriate delays and timings. I suggest, therefore, that for this example, “four” is pronounced with two syllables: White noise-low pitch-decreasing intensity-middle pitch. Then it goes digital. Each of these can be plugged into some logic gates to give unique results. I wish this bit could be analogue too: maybe it can be.

The digit, i.e. the output of the network of logic gates, is converted into a voltage of the appropriate level, ten times more with the second press, grouped into threes, so there are a series of currents but it doesn’t get ridiculously intense. An “overflow” is then added into the next set of circuits and reduced by a factor of ten. These are again stored somehow.

The next set of digits is stored in a second set of the same components. The “enter” button switches over to this set.

Finally, the user speaks one of the three words “add”, “subtract”, “multiply”, words of one, two or three syllables, easy to distinguish, and presses a second “enter” button. This analyses the words similarly and does one of the following:

Add: combines the voltages to produce an appropriate output. Subtract: switches the second voltage over to its negative equivalent. I have no idea if that’s feasible. That is then combined with the other voltage to produce a difference. Multiply: Adds repeatedly while reducing the other voltage until it reaches neutral.

Finally, somehow (again), this is turned into a series of pulses which drives a stepper motor which moves a stylus along a wax cylinder with a series of grooves. Each groove contains a recording of a single digit being spoken once. Each is played, then the stylus returns to the start, then the next digit is found in the same way and played. These recordings come out via a trumpet similar to that on a conventional record player.

So, to summarise, it goes: electronic analogue, electronic digital, electronic analogue, mechanical.

Could possibly be simplified by allowing a “Morse-code” style input and output and binary addition, subtraction and multiplication.

nineteenthly, May 18 2011

Fluidic logic http://en.wikipedia.org/wiki/Fluidics
[MaxwellBuchanan, May 19 2011]

Microfluidic logic http://pubs.rsc.org...ng/2007/LC/b708764k
[MaxwellBuchanan, May 19 2011]

Macrofluidic computations http://en.wikipedia...iki/MONIAC_Computer
Not boolean [mouseposture, May 20 2011]

Laminataed mouse brain http://www.technove...ntent.asp?Bnum=1073
[mouseposture, May 20 2011]


       By pitch, you are referring to vowel overtones?
pocmloc, May 18 2011

       Can you get it out with the beginning and the end being closer together?
MaxwellBuchanan, May 19 2011

       [MB] Try printing it out on a strip of paper, then taping the ends together. You could even apply a half-twist: that way, it would only be half a page in length.
mouseposture, May 19 2011

       I think that, in English, formants are more important than intonation. For instance, it is easy to understand numbers spoken in a monotone or a whisper, but difficult to interpret a hummed intonation pattern correctly.   

       This might be a better approach. You could distinguish formants using a small set of resonant circuits - or even Helmhotz resonators - feeding into a system of comparitors and latching circuits. You would need to speak in a robotic monotone, and it would probably need to be tuned to a single user, but in might be simpler to implement.   

       Using just the vowels (assuming the ability to detect two syllables and diphthongs correctly) the only ambiguity would be between 5 and 9. Saying 'niner' for 9 should fix that.   

       Wow, I think this could actually be done, with nothing more technologically advanced than a relay - which could be made using pre-industrial methods. In fact, it could even be purely mechanical, if you combined Helmholz resonators with vibrating membranes connected to fine clockwork escapements.
spidermother, May 19 2011

       //electronic digital// You could use analog registers for everything.
FlyingToaster, May 19 2011

       I tried the following simple experiment: Blow across a smallish bottle to find its resonant frequency. Pick a pitch that is a subharmonic of that frequency (e.g. a few octaves lower). Loudly vocalise a very slow vowel transition - e.g. 'waaii' drawn out over several seconds (just like when you are learning throat singing) towards the neck of the bottle.   

       I had to try a few combinations of bottle and pitch, but eventually got distinct resonance for some vowel sounds but not for others. It's not a great stretch to imagine a small set of bottles - probably about 4 or 5 - that produces a unique combination of resonances for each of the required vowel sounds.   

       The operator might also need to use slightly modified vowels, so the formants line up well with the harmonics, but skilled barbershop quartets do exactly that already.   

       As for detection, a quantity of conductive liquid in the bottom of each resonator, with a pair of contacts positioned just above the surface, might do.   

       [FT] Isn't //analog resister// a redundant pleonasm?
spidermother, May 19 2011

       Interesting. I'm annotating this only as a means of tracking it later if it becomes buried.
csea, May 19 2011

       [sm] possibly, if I knew what pleonasm was... and if you could spell "register" (which I did, I checked) :)   

       Resistors on the other hand... I imagine a register would just be a capacitor-based circuit.   

       Instead of resonant bottles (which sound nifty) you could go with a pitchfork (umm right word?) and a piezo crystal.   

       But either method requires the operator to be tuned up.
FlyingToaster, May 19 2011

       With the exception of the word "pleonasm", i'm sort of bobbing up and down at the top of an ocean of incomprehension here, but may at some point be able to reach the shore with some diligent textbook reading.   

       Sorry about the length. I seem to have lost the skill of brevity.   

       Concerning intonation, i mean that vowels are purer sounds than voiceless consonants, so the audio characteristics of a word should shift in that way, though not necessarily in pitch as such. I think vowels are chords. Certainly agree that "niner" and concentrating on vowels might be the way to go. I also wonder if some languages might have better words for numerals than others in this respect.   

       Very interested in knowing what a Helmholtz Resonator is. Right, done that, that's next Monday's session sorted! Anyway, what i'm wondering now is what would happen if the vocal tract sometimes has a distinctive shape for each number and if that shape would resonate when the appropriate word was spoken. It would be limited but might help to weed some out. I now envisage a series of moulded mouth and nose shapes linked together and a way of detecting resonance in the appropriate cavities in the cluster. It seems to me that would work for nasal consonants at least, meaning that "one" and "nine" could be distinguished from the others that way.
nineteenthly, May 19 2011

       //"register"// I read your annotation 3 times, and still misread it as "resister". Must have been thinking about electrical components.   

       //pitchfork// sp. tuning fork. A pitchfork is what your post-apocalypse compadres will wield when you show them your Magic Number Demon Machine.   

       //vowels are chords// Voiced (as distinct from whispered) vowels are similar to chords, in that there is more than one pitch present. However, since the fundamental is present, and the other pitches are harmonics of the fundamental, I would think of them as tones possessing a complex and distinct timbre, rather than proper chords.
spidermother, May 19 2011

       OK, got there. It's mainly about analog speech recognition for numbers, right?   

       I think that designing this system carefully is the wrong way to go. The way forward is to design it sloppily and then see what it does.   

       For instance, feed your input into a whole mess of poorly-thought-out crude frequency filters and what not. Accumulate the outputs of all the filters into capacitors, and have a threshold voltage at which each capacitor closes a switch to turn on a light. Maybe add a time dimension by having a stepper motor (better yet if you want to avoid digital: a large brass pendulum) redirect the output to different banks of filters as a function of time.   

       You'll quickly find that different vocalisations result in different patterns of lights coming on. There will be a lot of variation but also some consistency.   

       A series of photocells and photoresistors can then be used to detect chosen patterns of lights to represent each digit (for instance, lights 1,17, 9 and either 101 or 107 will result in a voltage coming out of the "FOUR" terminal, etc). It will all be very analoguey.
MaxwellBuchanan, May 19 2011

       Yes [MB], that's the twiddliest bit anyway. What you're saying, actually, makes me wonder if there's a way of designing this via an evolutionary approach, simulating random configurations until one works a bit and selecting that and so forth.   

       I like it.
nineteenthly, May 19 2011

       Now, of course, if you have a definite bent for the Steampunk side of things, and don't mind getting a bit wet, you could do all this by fluid logic. It seems a more natural way to process sound, somehow.
MaxwellBuchanan, May 19 2011

       Funny you should mention that, [MB], because i'm planning to waste even more time sticking a pneumatic-hydraulic alternate history on the Althist Wiki some time soon, and i thought of you.   

       Actually, while i'm at it, is microfluidics capable of logical functional completeness?
nineteenthly, May 19 2011

       This is brilliant, just my kind of thing and could be made to work.   

       I recently built what I loosely term the 'on board computer' for a bike project. The device consists of relays and a motorised mechanical sequencer. It's mounted on a board which consists of two layers of clear plastic, between which are sandwiched all the coloured wires, arranged orthogonally. It works reliably (so far) and is a pleasure to watch in operation.   

       I also like the fact that it would take at least as long to perform a calculation as a pocket calculator.   

       There's nothing wrong with electromechanical digital devices (think difference engine with solenoids). That would be my approach.   

Twizz, May 19 2011

       //is microfluidics capable of logical functional completeness?//   

       Well, for _micro_fluidics I'm not sure. There's effectively no inertia in microfluidics, so I'm not sure how much logic can be built with them.   

       However, most of the fluidic logic I've seen is larger in scale, and uses the inertia of the fluid to help control flows. I believe there is a full logic repertoire - at least I've seen AND, XOR, OR and probably some others, and I think you can build anything from those.   

       (Just as an aside and because I feel like showing off, I'm in the process of building a microfluidic chip with the connection ports modelled on an old Pentium III processor; they all connect to a microfluidic Sierpinski gasket in the middle. Totally useless but will look very cool.)
MaxwellBuchanan, May 19 2011

       Aha! A quick Google finds a paper implementing logic gates (at least AND and NAND, and maybe more) in microfluidic chips. However, the system doesn't look very robust. <link>
MaxwellBuchanan, May 19 2011

       //a full logic repertoire// Seems a waste of the technology. Surely a hydraulic computer would be best suited to solving differential equations <link>?
mouseposture, May 20 2011

       [Mouseposture], it could be AND rather than XOR.   

       Thanks for the reading [MB] and i like the gasket. Have you come across that video where video cameras are arranged to feedback to produce one on a screen? No digital processing there either.
nineteenthly, May 20 2011

       //Have you come across that video// link?
MaxwellBuchanan, May 20 2011

       Also, the use of rats might make this a lot easier. A single rat could probably learn a decent repertoire of sounds (perhaps processed, analogically, to bring their pitch more into a rat's comfort zone), and press levers accordingly. Some sort of Rat OS would need to be developed.
MaxwellBuchanan, May 20 2011

       That could happen. Also, their brains could be carefully extracted and networked.
nineteenthly, May 20 2011

       //their brains could be carefully extracted and networked.// <link>
mouseposture, May 20 2011

       (Obligatory) We'd have to get it out first. It'd have to be prepared. Treated. Diced.
spidermother, May 21 2011


back: main index

business  computer  culture  fashion  food  halfbakery  home  other  product  public  science  sport  vehicle