Horses make six recognized vocalizations: nicker, whinny, snort, blow, scream, and squeal. These are the only vocalizations that a pony should make (image at right compares a bio-horse whinny, “A”, with a human pony's (poor attempt at a) whinny, “B”).
Well, I decided to take this one step further. This will not be up everyone’s alley, but anyway. The scream, squeal and whinny are all relatively high frequency sounds with various modulations. I slightly modified a commercially available electronic bark collar such that it would deliver a shock for any vocalizations that were of a lower frequency than a typical whinny. It turns out that everyday speech is significantly below this frequency range, and it is quite difficult to produce words in the whinny frequency range.
The system is still being revised so that it will ultimately have a bandpass, allowing the lower frequency nickers and snorts as well as the high frequency whinny, scream and squeal. Further down the line I expect to advance the system such that it will not allow any attempts at high frequency speech (it's only somewhat effective at preventing that now).
Anyway, the piezoelectric transducer has to be on the neck to record the sounds, but it can easily be hidden (and/or integrated into) a collar. However, the shock does not have to be delivered to the neck area; it is trivial to connect a wire to the device which would deliver the shock to any part of the body (genital area seems the best).
The design is far from perfect since it relies on frequency ranges rather than composition so any high frequency sounds would be permitted. However, it is still in its very early stages. To illustrate this the image above right contains a spectrogram of one of my bio-horses whinnying (A) compared to a person (poorly) mimicking a whinny (B).
As you can see, the bio-horse whinny (A) has a much richer composition than the human version (B), which is by comparison just a narrow band with corresponding harmonics. For reference, normal human speech is 85-180 Hz for males and 165-255 Hz for females (from Wikipedia). The human whinny, even though it was of poor quality, did not result in a shock since its lowest frequency component was still well above that of normal human speech.
The collar I'm developing has quite a ways to go, however, I don't want to make it too restrictive since the whinny differs quite a bit from horse to horse. Nevertheless, it would be nice if it required more structure to a pony's whinny instead of allowing just any high frequency sound. Of course, the current collar is based on a modified commercial bark collar. I would need more sophisticated processing to get more discrimination.
Thus, my next step is to wirelessly link the collar to a phone or computer to perform the Fourier analysis and return the result to the collar. I will also be recording more whinnies from myself and from bio-horses to get a better feel for how discriminating I can/should be in determining what is a valid whinny for a pony. This is a longer term project, but I will post more information and source code as I get/develop it further.
Anyway, here is the current version (very crude, but it allows me to whinny but prevents me from talking):