What does an analog voice transmission look like in the visible spectrum? Analog radio signals are transmitted using light in the radio area of the spectrum.
If it was transmitted using the visible spectrum instead (using a visible light emitting device instead of a radio antenna, obviously), what would it look like?
Would it look like a solid light, because the vibrations would be imperceptibly fast? Or would one be able to visually tell when a voice started and stopped talking? Would you be able to see the difference between high and low pitches? Loud and quiet? Multiple tones and a single tone?
Has anyone ever done this (e.g. transmitted analog audio over fiber optic or something like that)? 
Edit: another way to think of this question. You know how you can take an x-ray video recording and then create a visual-light video with it (examples)? Imagine you do the same thing with a radio-wave video camera pointed at a radio tower broadcasting analog audio.
 A: Here is the spectrum of the human voice saying "oh":

If you translate the sound directly to light, adjusting the frequencies so they are visible by translating all the audible frequencies to all the visible spectrum, how would it look like? Well, it looks kind of similar to two of this:

that is the spectrum of a Wolf Rayet star. You can find more here. They are giant stars about to blow off as supernovae, and from afar they look pretty much like any other star: a white-blue-ish dot.
Now, a 2d spectrum of a sentence:

Each column in the plot is the spectrum of that part of the sentence. The qualitative properties look pretty much the same. Some parts have a bright spot in the higher frequencies range, that will be translated in a more blue tone, whereas others present a bulk in the lower frequencies, so they will look more red. In points like Fli there is signal all around, that reminds me of a black body, thus it will be mostly white.
I think the colours one would see are going to be somewhat similar to stars, where we can tell some have different tones, but it will probably not be blatantly obvious that they are. 
Edit
For the fun of it, here is a video of all this I am saying. The crappy code to generate is here, and I am naively using Colorpy to transform the spectrum into colour, that may not be accurate or correct. Actual sound may look lighter.
A: In this answer I will assume that we're doing amplitude modulation. That is , we have a light source at a constant frequency, and we use the audio signal to modulate its brightness. Unlike the encoding proposed in the other answer, this would be a straightforward circuit to build using an LED and an op-amp or a transistor or two. Although only a single frequency of light is used as the carrier signal, the whole frequency range of the audio signal can still be encoded. (AM radio uses the same principle.)
To a human eye, it would look pretty much like a continuous source of light, because the lowest frequencies in the human voice are higher than the highest flicker frequency that the human visual system can process.
To see this, note that an old-fashioned cathode ray tube television set effectively flashes on and off as the electron beam scans down the image. This happens at a frequency of 50Hz, but the resulting flicker is imperceptible to most people. 50Hz was chosen because it's close to the lowest frequency you can get away with - much lower than that and the flicker will become visible. (Movies are typically at 24 frames per second, but in that case the projector's shutter only has to close briefly in between frames, so it's a different situation.) According to Wikipedia, bass singers can typically only go as low as 80Hz, so if a voice signal were used to modulate a visible light source, the modulation would be essentially invisible to most human observers. 
Having said that, if you train yourself you can see the flicker of a cathode ray tube (or another pulsating light source such as a cheap LED lamp or a sodium street lamp) by rapidly moving your eyes past the light source and observing the afterimage it leaves on your retina. (At least, I've been able to do this since childhood; when I explain it to other people they seem to find it difficult to reproduce. It works best at night when other light sources won't interfere. It's also best if the light source is quite small or far away.) this shows that it's the brain and not the eyes that filters out the flicker. With this technique you can see frequencies up to maybe a few hundred Hertz, which would allow you to extract some information from the modulated light source. With training you could probably tell the difference between vowels and consonants, but I would guess not much beyond that. 
