This question comes from the context of two great works on popular science, one a book on cognition and artificial intelligence Gödel, Escher, Bach (GEB) And the other a fictional TV series Fringe.

In the series there is a particular episode, where a villain, kidnaps a mutant woman who "spontaneously combusts", from her house. Following this, the protagonist gets clues by analysing the partially molten glass of a window by cutting off a chunk and playing it like a record (TV show, so whatever fancy tech), finds out a very small bit of what the woman was saying while she was confronting her kidnapper.

In the book GEB, there is a particular dialogue which involves a massive record player which can play any tune on a single record by altering the record player, rather than the usual case where different arrangement of grooves give different tunes.

I have two related questions, one, is it possible to actually read voice info from the vibration patterns of partially molten glass; and two, is it guaranteed that the possible information will be a snippet of the actual conversation and not be random words/noise that happen to make "sense" due to possibly erroneous decoding?

Note: There is a relevant discussion in this wiki page on archaeoacoustics.
Note 2: For dialogue mentioned in GEB look at-

Achilles : What? A jukebox with only one record? That's a contradiction in terms. Why is the jukebox so big, then? Is its single record gigantic- twenty feet in diameter.
Tortoise : No, it's just a regular jukebox-style record.
Achilles : Now, Mr. Tortoise, you must be joshing me. After all, what kind of a jukebox is it that has only a single song?
Tortoise : Who said anything about a single song, Achilles?
Achilles : Every jukebox I've ever run into obeyed the fundamental jukebox- axiom, "One record, one song".
Tortoise : This jukebox is different, Achilles. The one record sits vertically suspended, and behind it there is a small but elaborate network of overhead rails, from which hang various record players. When you push a pair of buttons, such as B-1, that selects one of the record players. This triggers an automatic mechanism that starts the record player squeakily rolling along the rusty tracks. It gets shunted up alongside the record- then clicks into playing position.
Achilles : And then the record begins spinning and music comes out- right?
Tortoise : Not quite. The record stands still- it's the record player which rotates.
Achilles : I might have known. But how, if you have but one record to play, can you get more than one song out of this crazy contraption?
Tortoise : I myself asked the Crab that question. He merely suggested that I try it out. So I fished a quarter from my pocket, stuffed in the slot, and hit buttons B-1, then C-3, then B-10- all just at random.
Achilles : So phonograph B-1 came sliding down the rail, I suppose, and plugged itself into the vertical record, and began spinning?
Tortoise : Exactly. The music that came was quite agreeable, based on the famous old tune B-A-C-H, which I believe you remember...
Achilles : Could I ever forget it?
Tortoise : This was record player B-1. Then it finished, and was slowly rolled back into its hanging position, so that C-3 could be slid into position.
Achilles : Now don't tell me that C-3 played another song?
Tortoise : It did just that.


1 Answer 1


We will assume that the "molten glass" never actually became a liquid, as it would have fallen from the window and created a pool on the ground. The best we can hope for is that it became highly plastic, but still supported shear waves. Thus, we may treat the glass as though it were a normal pane of glass, but at some point it starts to become perfectly rigid, storing whatever displacement it had previously. This freezing must occur over a time scale much smaller than the smallest time scale that needs to be resolved. A useful upper frequency of intelligible speech is 3400 Hz, which corresponds to 290 $\mu$s. Thus, the freezing must be completed over a time of ~100 $\mu$s. Otherwise, the displacements will start to blend together, making any stored pattern useless. I am not an expert on glass cooling, but I am guessing it would be very difficult to cool it this quickly. Thus, even ignoring other potential complications (such as in-glass wave action [spoiler, it isn't a big effect]), I find it unlikely that sound could be stored naturally in glass as due to its cooling.

But, lets say that the vibration was magically frozen into the glass. Unlike a record, where the signal is stored in a very thin groove that is wrapped many times around the disk, this signal is stored over the entire window frame, and is highly redundant. In the best case scenario, the speaker will be located along the same wall as the window of interest, so that the signal can traverse the entire length of the window. Since the speed of sound in air is about 343 m/s, this means that to record a single second of audio would require a window that is 343 m wide. I don't know what size the windows in the show were, but they were probably closer to 2 m wide, which could record (2 m)/(343 m/s) $\approx$ 6 ms. So, I really doubt that a window could store a full conversation, and definitely could not be played like a record.

EDIT: There are many ways that the record could be extracted from the glass pane. Perhaps the most straightforward (in that you are reversing the manner in which it was made) would be to place a gramophone needle on the glass (imagine the glass pane flat on the ground so the needle is vertical) and then propel the gramophone needle along the glass at the speed of sound. (Alternatively, you could propel the glass at the speed of sound.) A more realistic method of obtaining the record would be to take a high-resolution image of the glass in a manner that gives a 3D rendering (I believe that you can do this with lidar, and almost certainly with multi-camera methods). Once this rendering has been obtained, use post-processing to pull out a line perpendicular to the ridges and valleys, which will be your signal. The final step is to determine the rate at which to replay your signal. If the speaker was along the wall of the window, the rate will be the speed of sound. If the signal is given by $Y(x)$, where $x$ is the position on the line, then the signal could be read as a true time series signal $Y(t)$, where $t=x/s$, where $s=c$ is the speed of sound. If the speaker was not perfectly in line with the window, then the angle with respect to the normal of the window will be used. Then $s=c/\cos(\theta)$, where $\theta$ is the grazing angle of incidence of the sound wave.

Note that as the angle approaches $\theta=\pi/2$ (perfectly normal incidence), the record speed approaches infinity. This record speed is also the speed that should be used above estimate how much signal can be recorded. If the speaker were directly in front of the window (and sufficiently far away that we can neglect the changing angle at the edges of the window), then the time that could be recorded would approach zero.

  • $\begingroup$ Nice answer...The conversation mentioned is about a second long from what I remember, and the "record player" involved an electron microscope of some sort, if it helps ;)!! $\endgroup$
    – Maan
    Dec 29, 2020 at 11:43
  • 3
    $\begingroup$ Well in very naive theory (with a lot of simplifications) one could do dinner scaling to move the same pattern in microscale and then "play it back" on different time scale pretty much like having the digital representation (samples) of a signal and arbitrarily choose the sampling rate (regarding the universal playback machine). As for the wall case... Pretty sure Michael M is right even with all those simplifications (the actual phenomena are very complicated) which means that it's a lot harder to practically happen. $\endgroup$
    – ZaellixA
    Dec 29, 2020 at 12:58
  • 2
    $\begingroup$ Perhaps the situation could be considered slightly less absurd (or significantly more, depending on how you look at it) if the speakers were traveling very quickly towards the window while they were speaking. If they were traveling close to the speed of sound and you ignore fluid dynamics, then you could compress the signal spatially (Doppler effect) to fit a full second onto a pane of glass. $\endgroup$
    – Michael M
    Dec 29, 2020 at 13:10
  • $\begingroup$ A quick calculation tells me that they would need to be traveling about 0.994 times the speed of sound, or about 763 miles per hour (341 m/s). $\endgroup$
    – Michael M
    Dec 29, 2020 at 13:14
  • 4
    $\begingroup$ I was just looking at the article on Archeoacoustics, and I found it quite fascinating. The key in the methods they use is some relatively constant force applied to a small area of a plastic surface in constant motion relative to the force. The examples I read were the smoothing blade on a wet pot being thrown and a paintbrush moving on wet paint. In order to recover the recording, the imprint speed must be matched, or else the spectrum will be distorted. While some guesswork is inevitable, speeding up or slowing down voice recordings does not significantly affect the intelligibility. $\endgroup$
    – Michael M
    Dec 29, 2020 at 13:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.