Historians of audio technology like to claim that Philipp Reis’s telephone of 1861 couldn’t have transmitted intelligible speech in the way he described it working, even if it might have done so in another way. But how would speech come out, I wondered, if it were sent over a system working exactly on the terms Reis proposed? I decided to try to find out through a digital simulation, expecting based on received wisdom to hear spoken language reduced to mere pitch contours and rhythms. The actual results came as a surprise. Here’s a sample. (CAUTION: I recommend you lower the volume before playing the audio examples in this blog post and then raise it to a comfortable level.)
Example 1 [download]
These are, to be sure, the results of an idealized simulation. I don’t mean to claim that any system built during the 1860s could have done quite so well, and I’m not in a position to judge the performance of any actual historical version of the Reis telephone with its particularities of materials and construction. But even so, I believe my results represent a legitimate test of the stated principle of Reis’s invention, as opposed to the stated principle of the Bell telephone. And that’s what interests me here: the viability of Reis’s conceptual approach to the transmission of speech, abstracted from any given physical implementation of it. Based on the results, I feel justified in concluding that speech transmitted just as Reis proposed can sometimes be intelligible but won’t always be intelligible. In other words, Reis’s method can’t mediate speech as well as Bell’s method can, nor is it incapable of mediating recognizable words at all; instead, it falls somewhere in between. That conclusion is entirely consistent with the earwitness reports we can read about its performances during the 1860s or early 1870s, as well as with research on the intelligibility of distorted speech published since the 1940s. It just happens to be at odds with how historians of audio technology have characterized matters for the past 140 years or so—including me, not so very long ago.
Reis and his contemporaries described his telephone as working according to a very specific and distinctive logic. First, at the transmitting station, the movements of a membrane in response to a sound wave rapidly opened and closed an electrical circuit. Then, at the receiving station, the resulting interruptions and restorations of current generated another sound wave, most often (but not always) based on the principle of “galvanic music,” in which the interruption of current in a coil was understood to produce a sound through magnetostriction in its core. During the late 1870s and 1880s, experts stated again and again that a telephone that worked on this principle could transmit only musical notes, and not intelligible speech.
Understanding that sounds can be characterized in terms of pitch, loudness, and quality, they considered how a sound wave (shown above in blue) would be transduced by Reis’s transmitter into an electrical signal (shown above in red), and concluded that the latter could only embody pitch (i.e., frequency), whereas sound quality or timbre was understood as depending on the wave shape and would therefore be lost. William Henry Preece accordingly told the British Association in 1877:
It is evident that in Reis’ telephone, everything at the receiving end remained the same excepting the number of vibrations [per second], and therefore the sounds emitted by it varied only in tone, and were therefore notes, and nothing more. The instrument remained a pretty philosophical toy, and was of no practical value.
Judge John Lowell further stated in an oft-cited circuit court decision of 1881 that Reis’s telephone had been based on “a false theory”; that “articulate speech could not be sent and received by it”; that this “deficiency was inherent in the principle of the machine”; and that it had therefore not meaningfully anticipated Bell’s telephone as an instrument for mediating the spoken word by electricity. “A century of Reis,” he proclaimed, “would never have produced a speaking telephone by mere improvement in construction.”
And yet numerous people claimed that they had been able to recognize spoken words transmitted by Reis’s telephone, to wit:
- “At an experiment which we made at Friedrichsdorf, in the presence of Hofrath Dr. Müller, Apothecary Müller, and Professor Dr. Schenk, formerly Director of Garnier’s Institute, an incident occurred which will interest you. Singing was at first tried; and afterwards his [i.e., Reis’s] brother-in-law, Philipp Schmidt, read long sentences from Spiess’s ‘Turnbuch’ (Book of Gymnastics), which sentences Philipp Reis, who was listening, understood perfectly, and repeated to us. I said to him, ‘Philipp, you know that whole book by heart;’ and I was unwilling to believe that his experiment could be so successful unless he would repeat for me the sentences which I would give him. So I then went up into the room where stood the telephone, and purposely uttered some nonsensical sentences, for instance: ‘Die Sonne ist von Kupfer’ (The sun is made of copper), which Reis understood as, ‘Die Sonne ist von Zucker’ (The sun is made of sugar); ‘Das Pferd frisst keinen Gurkensalat’ (The horse eats no cucumber-salad); which Reis understood as ‘Das Pferd frisst….’ (The horse eats…).” (Heinrich Friedrich Peter, Thompson, p. 127)
- “A good instrument reproduced the words sung into it in such a manner that not only the pitch but also the words of the song were perfectly understood, even when the listener was unacquainted with the song and the words.” (E. Albert, Thompson, p. 45)
- “There is not a shadow of a doubt about Reis having actually achieved imperfect articulation. I personally recollect this very distinctly, and could find you plenty more people who witnessed the same.” (Rudolph Messel, Thompson, p. 46)
- “I listened [in 1864] at the latter part of the apparatus, and heard distinctly both singing and talking. I distinctly remember having heard the words of the German poem, ‘Ach! du lieber Augustin, Alles ist hin!’ &c.” (G. Quincke, Thompson, p. 113)
- “I myself spent considerable time with him [Reis] in transmitting words through the instruments. We never (in my time) got the length of transmitting complete sentences successfully, but certain words, such as ‘Wer da?‘ ‘gewiss,’ ‘warm,’ ‘kalt,’ were undoubtedly transmitted without previous arrangement.” (Ernest Horckheimer, Thompson, p. 117)
- “Before disposing of the apparatus, I showed it at the November meeting (1865) of the Dublin Philosophical Society, when both singing and the distinct articulation of several words were heard through it, and the difference between the speakers’ voices clearly recognised.” (S. M. Yeates, Thompson, p. 47)
- Q. “At the time of your experiment [with a Reis telephone] in Room 24, January 7th, 1869, was anything said by anyone there present, with respect to the hearing the words of the song? And if so, what and by whom?” A. [after an objection] “There were about half a dozen persons present, who said they could understand the words of the song; among them Mr. Hatch and Mr. Finnell. They told me so….” (Deposition of Peter Van der Weyde, 1884, p. 12)
How could this be? Experts soon settled on an explanation that was consistent with their late nineteenth-century understanding of telephony and why and how it worked. If the transmitting membrane of the Reis telephone hadn’t fully opened and closed a circuit, but had only varied its resistance via the degree of contact between two points, then it could have transduced sound waves into a continuously variable current just as later telephones did, and everyone knew by then that it was possible to transmit intelligible speech like that. This alternative mode of operation would have run contrary to contemporaneous descriptions by Reis and others, but it could have been achieved by mistake if the device had been put “out of adjustment.” Some commentators went further, arguing that Reis’s theoretical remarks about sound curves and other clues implied that he must have intended to vary the degree of contact all along (a position with which I disagree). But perhaps the most compelling evidence of all was the empirical witness testimony about Reis’s telephone having in fact transmitted intelligible words. If one accepts that a telephone working on the make-and-break principle can’t transmit intelligible speech, but that Reis’s telephone had transmitted intelligible speech, then it follows that it must have done so on some other basis—presumably a loose contact. And if Reis’s telephone had transmitted speech only sporadically and imperfectly, that could always be chalked up to ignorance about the conditions that sometimes allowed it to succeed. Silvanus Thompson promoted this view in his exhaustively researched biography of Reis (1883). More recently, in 2003, news media reported that an experiment at the Science Museum in London back in 1947 had confirmed that Reis’s instruments could transmit speech perfectly well if the signal were suitably amplified, but that these results had been suppressed so as not to embarrass the Bell interests—a data point that received further exposure through inclusion in Seth Shulman’s The Telephone Gambit (2009, at p. 125). I imagine the transmitter in that experiment had been purposely adjusted so as to present a loose contact, as had been done in a similar experiment of 1932 that wasn’t “hushed up” and so hasn’t received nearly as much attention from conspiracy aficionados (see relevant sources quoted by Basilio Catania).
It may well be that Reis’s telephone sometimes transmitted continuously variable signals using a loose contact during the 1860s—my purpose here isn’t to dispute that claim. Rather, I want to challenge the common assertion among telephone historians that a membrane controlling a simple on-off switch can’t transmit intelligible speech; in other words, that Reis’s so-called “false theory” was indeed false, as far as it went. If that claim isn’t true, then the earwitness testimony cited above shouldn’t be taken as evidence of Reis’s use of a loose contact, whether intentional or not; and historians should stop equating the “speaking” telephone straightforwardly with instruments using variable resistance.
Back in July 2017, I tried an experiment where I took an existing digital sound file and converted it into one-bit audio, equivalent to a simple on/off signal. This is easy enough to do. If our range of sample values runs from -1 to 1, we can write a little code—as I did—to reassign each sample greater than zero to 1 (“on”) and all other samples to -1 (“off”). Alternatively, anyone who isn’t comfortable with coding could just boost the gain in standard sound editing software until it becomes “infinitely clipped.” It turns out that when we do this with a recording of speech, the results are quite intelligible, as illustrated by the following excerpt I’ve processed from the soundtrack of the British Pathé film “The Talking Link” (1936).
Example 2 [download]
I was pretty excited about this “discovery” until I learned through some quick googling that the intelligibility of one-bit speech is common knowledge—just not, apparently, among telephone historians. I had only reinvented the wheel, so to speak. Here, for example, was an acknowledgment I found tucked away in a footnote by Manfred Schroeder in his 2004 book Computer Speech (p. 28):
As early as 1860 Philip Reis, a German professor, had constructed a premature telephone, which however was not officially recognized as capable of transmitting intelligible speech because it was basically an on-off switch activated by sound waves. Of course, we now know that one-bit (“infinitely clipped”) speech can be quite intelligible, but a hundred years ago Reis did not prevail with his binary speech signal.
To the best of my knowledge, Schroeder was the only person who had previously drawn a connection between Reis’s work and the known intelligibility of one-bit speech, and he was an acoustician and computer scientist, not a telephone historian. Still, the latter discovery has been a well-documented part of the scientific record since the 1940s. I’ll discuss the earlier research into one-bit audio and its implications for Reis in more detail below.
But the process I’ve described so far isn’t yet a fair simulation of the Reis telephone. The amplitude values in a WAV file conventionally represent velocity, i.e., how fast the membrane in a microphone or speaker moves backwards or forwards. But in the Reis telephone, it was the displacement of the membrane past a given spatial threshold, corresponding to its resting position, that interrupted or restored the current. To convert the velocity data in a WAV file into displacement data, we need to take its cumulative sum, sample by sample; and then it’s necessary for practical reasons to apply a high-pass filter afterwards to remove DC offset (at, say, 20 Hz). If we now take the resulting samples corresponding to displacement and assign those greater than zero to 1 (“on”) and those less than zero to 0 (“off”), we’ll be simulating the way in which a Reis transmitter would have processed the membrane vibrations into an electrical signal. This gives us a lopsided-looking signal with no negative values, but that’s arguably what we want.
Our binary signal still represents displacement, insofar as it represents anything at all; so if we want to convert it back to a representation of velocity—the parameter which modern headphones or loudspeakers are designed to transduce into sound waves—we need to calculate the differences between each pair of samples, reversing the process of taking a cumulative sum which we carried out before. This approach will “reproduce” the binary position of the transmitting membrane in the receiving membrane of our speaker, causing the latter to move instantaneously (in theory, but in practice as close to this as is mechanically feasible) from one extreme in its range of movement to the other with each interruption or restoration of current. It’s arguably the most faithful way we can convert a Reis transmitter signal into sound waves according to modern audio logic.
Example 3 [download]
But this approach doesn’t necessarily match what Reis receivers originally did with this kind of signal.
Reis used two main types of receivers, electromagnetic-tympanic and magnetostrictive. The magnetostrictive receiver was standard, whereas electromagnetic-tympanic receivers were described only rarely (in Wilhelm von Legat’s 1862 article in the Zeitschrift des deutsch-österreichischen Telegraphen-Vereins, as well as in an article in Neues Frankfurter Museum published in October 1861 which I speculate may also have been written by von Legat).
With an electromagnetic-tympanic receiver, whenever the current was on, an electromagnet would have tugged the receiving membrane in one direction; but whenever the current was off, the membrane would have been free to return to its resting position. With the former movement, the electromagnet would presumably have exerted a constant force on the membrane the whole time it was active, though its actual effect on the membrane might have varied over time: as its displacement increased, I imagine it would have put up more resistance to moving further, since it couldn’t have kept stretching indefinitely. However, a modern speaker should have comparable limitations. The latter movement back towards a rest position would, I guess, have depended entirely on the elastic properties of the membrane itself; again, a modern speaker should behave comparably. In neither case would the receiving membrane have moved instantaneously from extreme to extreme, or even done anything close to this. It may, on the other hand, have undergone nearly instantaneous changes in velocity, subject to complications of coil inductance and resistance, membrane mechanics, and so forth.
The simplest way to model what I’ve just described—ignoring the aforesaid complications—is to say that our receiving membrane should have a constant velocity in one direction as long as the current is “on” (being tugged continuously forward by the electromagnet) but should be left alone to return towards its rest position based on its own elasticity while the current is “off.” The result is that we end up treating the Reis transmitter signal, which actually represents the displacement of the transmitting membrane, as representing velocity in the receiving membrane. In other words, we pass the Reis on-and-off signal to the headphones or loudspeaker as is, without first taking the difference between samples. The perceptible effect on the resulting sound is to boost lower frequencies at 6 dB per octave.
Example 4 [download]
Meanwhile, what about a magnetostrictive receiver? Magnetostriction causes a change of shape proportional to strength of magnetization, transducing magnetic energy into mechanical energy (or vice versa). In a make-and-break system, the shape would presumably undergo a change with magnetization and another change with demagnetization, so I suspect the behavior might be closer to that modeled in Example 3 above. According to Charles Grafton Page’s findings, as published in 1837 in the American Journal of Science and Arts:
The ringing [of an electromagnet] is heard both when the contact is made and broken; when the contact is made, the sound emitted is very feeble; when broken, it may be heard at two or three feet distance…. [T]he sounds produced [by electromagnets of different sizes] differed from each other, and were the notes or pitches peculiar to the several magnets. If a large magnet supported by the bend be struck with the knuckle, it gives a musical note; if it be slightly tapped with the finger nail, it returns two sounds, one its proper musical pitch, and another an octave above this, which last is the note given in the experiment.
In Germany, G. Kessler-Gontard apparently only noticed the louder sound made when the current was interrupted, judging from an account of his parallel discovery published in 1840 in the Archiv der Pharmacie. In the attached file, I’ve tried to model this by assigning each amplitude transition from positive to negative the value -1 and each transition from negative to positive the value 0.1 (to reflect its comparative weakness).
Example 5 [download]
This last example uses a single-sample “tick” to represent a sound characterized in the nineteenth century as a “magnetic tick”; see e.g., Robert M. Ferguson, Electricity (1866), at p. 182. I’m not honestly sure how defensible that is. But the audible result doesn’t seem that much different from the previous two as far as intelligibility goes, albeit differently balanced. Meanwhile, there are in fact magnetostrictive receivers out there which can be listened to and evaluated. One particularly nice-looking example with a nickel core and wooden housing, built by Simplifier, can be seen in action in a YouTube video. Substituting a core made of the modern alloy Terfenol-D, with much higher magnetostriction, can improve the response yet further; and, indeed, that’s the material that was used in the Olympia Soundbug, a magnetostrictive receiver introduced commercially in 2002, and the follow-up Whispering Window. Reviewers often described the Soundbug as sounding “tinny,” and I think the Simplifier transducer could be described similarly (I wonder in passing whether this might follow from differentiation with its 6 dB/octave slope, and whether integrating the signal beforehand might improve performance). But in general, the differences in behavior between electromagnetic-tympanic and magnetostrictive receivers appear unlikely to have led to any fundamental qualitative difference affecting intelligibility.
I’ve disregarded many factors in my simulations, including the resonances of membranes and iron cores, the frequency responses of membranes used to pick up sounds, and the effects of inductance (although this last might just mitigate the noise occasioned by square waves). If we want to address such factors, the next step would probably be to build and experiment with some working replicas of Reis telephones, taking care to operate them in make-and-break mode—something I suspect nobody may have had a motive to try since the 1870s. After all, if we assume from the outset that a Reis telephone operated in make-and-break mode can transmit only musical pitches, what would be the point in replicating that? (A hybrid scenario also occurs to me: namely, even if a Reis transmitter fully breaks a circuit during part of each vibration cycle, it could also vary the degree of contact during the part of the cycle when the circuit is closed. If the threshold is set at the zero crossing, the wave shape of half of each cycle could be transduced.)
But as I stated earlier, my present focus is the so-called “false theory” itself. The question on the table is not whether Reis’s telephone worked, but whether it’s possible in theory to transmit intelligible speech by causing the vibrations of a membrane to interrupt a current of otherwise consistent voltage. The answer appears to be “sort of,” which differs as much from “no” as it does from “yes.”
The intelligibility of one-bit audio was first reported in two articles: J. C. R. Licklider, Dalbir Bindra, and Irwin Pollack, “The Intelligibility of Rectangular Speech-Waves,” American Journal of Psychology 61:1 (January 1948): 1-20; and J. C. R. Licklider and Irwin Pollack, “Effects of differentiation, integration and infinite peak clipping upon the intelligibility of speech,” Journal of the Acoustic Society of America 20 (January 1948): 42-51. J. C. R. Licklider is remembered today for his pioneering contributions to interactive and networked computing. His two coauthors were Ph.D. students at the time, but both went on to have distinguished careers, with Irwin Pollack known for his innovative research on the auditory system and Dalbir Bindra for his work on the neurophysiology of motivation and fear. Licklider, Bindra, and Pollack had begun by studying the effect on speech intelligibility of various amounts of clipping—that is, limiting or “cutting off” the peaks and troughs of a waveform when they pass a given threshold. As they stated in the first-mentioned article (p. 4):
To those of us who had assumed that faithful reproduction of the speech-wave was essential for high intelligibility, the results of the initial experiments were quite surprising. Never did the articulation scores fall below 96%. Insofar as intelligibility was concerned, communication was essentially perfect—just as much so with nine-tenths of the speech-wave clipped off as with high-fidelity reproduction of the entire wave.
When they proceeded to experiment with “infinite” clipping, they found that it “causes no great difficulty and makes necessary only few repeats in ordinary conversation” (p. 6).
Figure from Licklider, Bindra, and Pollack (1948), at p. 7.
In the second-cited article, Licklider and Pollack—this time writing without Bindra—presented the results of experiments in which an input signal had been differentiated (i.e., converted into its rate of change), integrated (i.e., converted into its cumulative sum), or left unaltered before infinite clipping. They found that the unaltered signal was 85.9% intelligible after clipping; the differentiated signal 97.9% intelligible; and the integrated signal 14.8% intelligible (Table IV, p. 50). By contrast, integrating or differentiating after clipping made scarcely any difference.
Which of the three signal types subjected to clipping corresponds best to the behavior of a Reis transmitter? I’ve found that surprisingly hard to answer, although it may be obvious to others with more expertise than I have. At first, I assumed the unaltered signal in these experiments would have resembled a modern audio signal, with voltage proportional to velocity, in which case the integrated signal should correspond to displacement as in a Reis transmitter. However, Licklider and Pollack had recorded their test words onto “acetate” (i.e., lacquer) discs at a time when the standard pickup for playback would have been piezoelectric, with an electrical output proportional to displacement, not velocity. That circumstance would instead seem to point to the unaltered variant corresponding to the Reis transmitter signal. So which is it?
Faced with this uncertainty, I decided to try carrying out a similar experiment digitally and to compare my results with those of Licklider and Pollack. Here’s a short speech recording with infinite clipping applied to (1) an integrated displacement signal, (2) a displacement signal like that of a Reis transmitter, (3) a velocity signal equivalent to a differentiated displacement signal, and (4) a differentiated velocity signal. Each example is differentiated relative to the one that comes before and integrated relative to the one that comes after.
[download]
I find a big jump in intelligibility between (1) and (2), with another major improvement in (3), but little if any additional improvement in (4). Judging from this result, I suspect that (2) most likely corresponds to the signal unaltered before clipping in the 1940s experiments, (1) to the integrated signal, and (3) to the differentiated signal. It’s true that (1) seems far less than even 14.8% intelligible, but Licklider and Pollack acknowledge that their test subjects had learned which words were on the 250-word test list over time, thereby gaining the ability to recognize words based on cues that wouldn’t ordinarily have been sufficient for comprehension (pp. 46-7). Moreover, according to a graph (Fig. 4 on p. 46), subjects had managed to identify only about 3% of words correctly during their first test session with speech integrated before clipping, which was itself probably not their first test session overall, and hence not their first exposure to the word-list. For comparison, they had managed to identify just over 50% of words during their first test session with speech unaltered before clipping, and around 74% the next time around, before going on to achieve scores averaging around 90% by the end of the project and 85.9% as an overall cumulative score. That seems about right for the Reis transmitter simulations I’ve prepared. This interpretation would also suggest that a standard modern audio signal—with voltage proportional to velocity—would correspond to the differentiated source in the 1940s experiments, such that differentiating before clipping today would no longer confer the advantage it did then.
The question of why infinitely clipped speech is intelligible has received some further attention since then. Arthur L. Fawe, “Interpretation of Infinitely Clipped Speech Properties,” IEEE Transactions on Audio and Electroacoustics AU-14:4 (December 1966):178-183, aims to come to terms with it theoretically, going way over my head in the process. Daniel Kahn, “Cues in the perception of infinitely clipped speech,” Journal of the Acoustical Society of America 78 (1985), S49, provides empirical evidence that the explanation lies in the power spectrum rather than in zero crossing locations. I don’t get the sense that the issue has yet been fully resolved. As for the effects on intelligibility of integration or differentiation before clipping, however, I imagine those can be explained rather simply in terms of frequency balance. Speech sounds can presumably be recognized best in those cases where the fundamental and one or more upper harmonics are similar enough in intensity to compete with each other for control of zero crossings, resulting in rapid alternation between or among frequencies on the receiving end. Integration boosts lower frequencies, making it less likely that upper harmonics will overpower fundamentals; while differentiation boosts higher frequencies, making it more likely that upper harmonics will overpower fundamentals. If this explanation is correct, then the intelligibility of speech transmission by Reis’s method could be expected to vary widely depending on the interaction of the frequency response curve of a given pickup membrane with the harmonic structure of a given speaker’s voice.
Meanwhile, Reis’s staunchest defenders have tended to be the most eager to downplay the importance of the so-called “false theory” in understanding and assessing his work. Consider what Silvanus Thompson has to say about the matter in his Reis biography (at pp. 132-3):
It has often been said, but incorrectly, that Reis intended his “interruptors” or contact regulators to make and break the electric circuit abruptly in the manner of a telegraphic key worked by hand. No doubt in the mouth of a professional telegraph operator the words “interrupting” the circuit, and “opening” and “closing” the circuit, do now-a-days receive this narrow technical meaning. But Reis was not a professional telegraph operator: he did not…even know the signals of the Morse code, and it is self-evident that he did not use the terms in any such restricted or unnatural sense as abrupt “make-and-break,” because he proposed at the outset to interrupt the current in a manner represented by the gradual rise and fall of a curve, stating emphatically in his very first memoir on telephony…, that to reproduce any tone or combination of tones all that was necessary was “to set up vibrations whose curves are like those” of the given tone or combination of tones.
I disagree. It’s disingenuous to ascribe Reis’s references to “interrupting,” “opening,” and “closing” a circuit to a terminological clumsiness that concealed his real thoughts—to claim that when he wrote “interrupting,” what he really meant was “varying.” To all appearances, Reis was quite capable of expressing his ideas clearly, so far as they went; and closer investigation reveals that the principle of making and breaking a circuit was essential to a consistent line of reasoning that extends throughout all facets of his project. Consider:
- In Reis’s published article on the telephone, he wrote that the vibrations of the eardrum “induce an equally rapid lifting-up and falling-down of the hammer on the anvil (according to others: approach and recession of the atoms of the ossicles) and an equally great number of concussions of the cochlear fluid in which the auditory nerve spreads out with its ends. The greater the compression of the sound-conducting medium in a given moment, the greater the vibrational amplitude of the membrane and the hammer, consequently the more powerful the strike upon the anvil and the concussion of the nerves through the mediation of the fluid” (my own translation from the German, emphasis added). Judging from this passage, Reis understood the ear itself as able to pick up only discrete “concussions” of variable intensity, and not a continuously variable wave shape, even if he acknowledged in parentheses that “others” might think otherwise.
- Reis wrote further: “The strength of this [reproduced] sound…stands in proportion to the original sound, for the stronger the latter is, the greater are the movements of the eardrum, the greater the movement of the hammer, the greater the period of time during which the circuit remains open, and consequently the greater—up to a certain limit—the movement of the atoms in the reproduction-wire, which we perceive as a greater vibration, just as we would have perceived the original wave” (again, my own translation from the German, emphasis added). Here we find Reis arguing that a make-and-break system would mediate variations in intensity as a consequence of mediating variations in duration.
- The “curves” Reis illustrated in the body of his article consist of discrete sinusoidal impulses at different intensities, consistent with individual “strikes” of hammer against anvil.
Reis suggested that such patterns in intensity might be responsible for different vowel sounds, which he knew past researchers had managed to synthesize using toothed cogs. As we’ve seen, he also thought his system could mediate a signal patterned in this way via variable duration of contact. On the other hand, faced with explaining why his telephone seemed to be able to transmit consonants better than vowels, he speculated that the reduction in the absolute amplitude of reproduced vowel patterns might be responsible. - Reis’s article also included three fold-out plates of more nuanced compound wave shapes corresponding to combinations of musical notes. However, he referenced these plates before proceeding to the arguments summarized above, so I believe we have to assume he didn’t perceive them as contradicting the rest of his scheme. According to an article published in the Neues Frankfurter Museum of November 3, 1861, Reis had come up with a distinctive “theory of auditory sensations” according to which “our ear actually never perceives several notes at the same time (any more than the eye perceives several colors at the same point in the field of vision), but always perceives only one note, though perhaps peculiarly modified by the interaction of different waves, and through training acquires the ability to discover again the elementary wavelengths that were necessary for that peculiar modification” (see my translation of the whole article here). As evidence in support of Reis’s theory, the author—who I believe may have been Wilhelm von Legat—pointed to the recognizability of transmitted piano chords. The picture that emerges hazily from these sources is one in which Reis believed the ear was only physiologically capable of picking up discrete concussions of varying intensity, but that people somehow learned to reconstruct sound spectra from them.
- As far as Reis and his contemporaries would have been aware from reading the work of Charles Grafton Page and others, the production of sounds via magnetostriction requires an interruption of current. The fact that we know differently today can’t be held to explain Reis’s thought processes back in 1861.
This analysis is one I came up with independently, but I acknowledge that John E. Kingsbury anticipated parts of it in The Telephone and Telephone Exchanges (1915), at pp. 125-139—an older piece of Reis scholarship that provides a nice counterpoint to Thompson’s. As far as Kingsbury was concerned, of course, he was making a case for why Reis’s telephone wouldn’t have been able to transmit speech, which may be why it seems so rarely to have found its way onto the reading lists of Reis apologists.
It’s been a matter of general consensus since around Reis’s time that the ear can pick up multiple frequencies simultaneously. But the fact that Reis seems to have been mistaken about how the ear works doesn’t mean his views about hearing couldn’t have inspired him to design a partially-successful speaking telephone. After all, Thomas Edison—who took Reis’s work as a starting point—was operating with a virtually identical hypothesis about the acoustics of speech when he devised both his carbon button transmitter and his speaking phonograph (see here and here for details). Theories don’t need to be “correct” in order to inform successful inventions. And, in any case, the claim about Reis’s telephone being based on a “false theory” involves the make-and-break approach itself, and not the ideas that led Reis to adopt it.
Even if Reis’s make-and-break approach can transmit somewhat intelligible speech, the invention of a variable-resistance telephone capable of transmitting entirely intelligible speech—not at first, certainly, but with sufficient development—would still have counted as a major improvement on it, deserving of patent protection in its own right. For that reason, I doubt any points I’ve raised here would have materially affected the outcome of any nineteenth-century telephone litigation, even if some of the decisions might have been worded differently (e.g., “A century of Reis would never have produced a fully and consistently intelligible speaking telephone by mere improvement in construction”).
Commentators in the past have struggled to characterize the audible effects of the make-and-break approach on a speech signal. Consider A. Edward Everson’s attempt at a description in The Telephone Patent Controversy of 1876 (2000), at p. 157:
The main problem with the Reis telephone was that it could operate in either of two modes: (1) as a make-and-break transmitter, or (2) as a true microphone, or non-make-and-break instrument. Whereas the former mode could transmit sounds, such as musical notes, its speech-transmitting ability was quite limited because the opening and closing of the contacts would distort or remove the voice’s vocal inflections. It gave a monotone quality to the reproduced sound. However, the microphonic mode, which soon became the only telephone transmitter mode, could transmit the voice in all its tones and inflections. In microphone mode, the contacts are neither fully closed nor fully open.
In fact, it’s the inflections—in the sense of pitch contours—which the Reis telephone was best equipped to mediate when operating in make-and-break mode, as even authorities of the 1880s recognized. According to received wisdom, the reproduced sound should have been monotimbral, not monotonous. In fact, my simulations suggest it wouldn’t necessarily have been monotimbral either. It would have been distorted, to be sure, but that observation isn’t very enlightening, since distortion can take many forms. In the end, I’m not sure how best to express the effects on sound quality in words, so it’s fortunate that the sound files can speak for themselves.
Can we know for sure whether Reis’s telephone was working in make-and-break or loose-contact mode back in the 1860s when earwitnesses claim it transmitted recognizable words? Maybe. Here’s an excerpt from the biography of Philipp Reis at Britannica.com, which I believe dates back to the print edition of 1990:
[T]here were several reports of successful speech transmission. These reports were subsequently discounted in court cases upholding the patents of Alexander Graham Bell, largely because it was recognized that speech transmission would have been impossible if the instruments had operated as Reis believed they did. Nevertheless, it is a fact that, if the sound entering a Reis transmitter is not too strong, contact between the metal point and the metal strip will not be broken. Instead, the pressure of the former on the latter will fluctuate with the sound, causing fluctuations in the electrical resistance and therefore in the current. Similarly, the receiver will respond to continuously fluctuating as well as to intermittent currents (but not by magnetostriction [sic; the transduction is still based on the core’s change of shape proportional to strength of magnetic field, albeit not on the “ticks” of “galvanic music” occasioned by interruptions and restorations of current]). The sensitivity, however, is extremely low—so low that it is not unreasonable to question the validity of the limited testimony regarding successful voice transmission in the 1860s.
The author of this entry doubts whether a magnetostrictive Reis receiver could have transduced a continuously variable signal into sound at high enough volume for it to be heard by a lecture-hall audience. It will be recalled that the experimenters at the Science Museum in 1947 had to amplify the signal to get audible results in this way. By contrast, interrupting the current would have produced reasonably loud sounds in a magnetostrictive Reis receiver. Thus, if we accept that it is in fact possible to transmit recognizable words by making and breaking a circuit, it strikes me as far more likely—based on sheer audibility—that this is what was happening back in 1861.
I have one last data point to share. When Heinrich Friedrich Peter spoke the words “Das Pferd frisst keinen Gurkensalat” into a telephone, Philipp Reis reportedly understood only “Das Pferd frisst….” at the other end of the line, without being able to make out the rest of the sentence. I found a recording of an announcer speaking these words in a radio program about Reis (“Philipp Reis—Erfinder des Telefons” on Radiowissen) and—wholly without permission (sorry)—ran it through the same processing algorithm I used for Example 4 above. Here’s what my simulation suggests Reis might have heard.
[download]
I would like to express my thanks to Christie Crews of Wellington, Kansas, for her delightful LibriVox readings from The Peter Patter Book of Nursery Rhymes, which provided the source audio for many of my examples. Her voice is uncommonly well suited to Reisian processing. Thanks also to David Giovannoni for his thoughtful feedback on many of the ideas presented here. Of course, responsibility for any and all harebrained misstatements rests with me alone.
PS. (December 20, 2020): Since posting the above, I’ve run across a possibly relevant article by Ja Hyon Ku (구자현) entitled “The Invention of Reis Telephone and Its Problem of Speech Quality,” published in the Journal of the Acoustical Society of Korea 29:6 (2010): 395-401. The abstract is in English and mentions that Reis “used the intermittent electricity in accordance with the experimental tradition in European acoustics, occasioning the speech quality of his telephone to have a fatal shortcoming.” The article itself is in Korean, which I regrettably can’t read. I note that the bibliography doesn’t cite the canonical works on the intelligibility of one-bit audio, so I’m guessing that doesn’t come up. However, Ku’s figure 3 contains a nice graphical analysis of the three steps in Reis’s signal chain (displacement of transmitting membrane, intermittent current, displacement of receiver), suggesting that he had worked through that important piece of the puzzle. Ku’s main goal seems to be to account for why Reis’s telephone wasn’t as good as Bell’s, but I’m curious what exactly he has to say about the “fatal shortcoming” of its “speech quality” and would be grateful to anyone who’s in a position to summarize.
Griffonage-Dot-Com 