New Directions in Phonautographic History

Patrick Feaster, Indiana University
Presented at the ARSC Conference in Washington DC, May 29, 2009

(may take a while to load)

The edited sound file hosted by First Sounds is based on the raw recording of my presentation available from ARSC.  I’ve interrupted the former at several points where I ran into technical difficulties, but otherwise it’s as it was, even unto my consistent mispronunciation of “Au Clair de la Lune." I've included select stills from my original PowerPoint, though I've left some images out pending official publication. The text below is slightly modified in light of omitted visuals, animations, etc.

This morning, I’m excited to be able to retell the story of Édouard-Léon Scott de Martinville’s efforts to record sound in light of the newly discovered source materials David has just reported to you.  Here’s an alternative portrait of Scott I learned about just last month, published in the Phono-Gazette of May 1, 1905—only the second picture of him I’ve seen.  Judging from the hair, this is Édouard-Léon at a younger age, probably a lot closer in time to his invention of the phonautograph—he was forty years old in 1857.  It’s still the same man—the same face—but from a different angle, and combined with the earlier picture, we start to get a better sense of depth, even of personality.  The other unexpected source materials now at our disposal likewise give us a much richer picture of Scott—both the man and his work—than we had before.  For the first time, we’re in a position not just to prove that phonautograms can be played back—as neat as that is—but to use those sounds in turn to understand and illustrate the whole course of Scott’s work.

Until just a few months ago, my First Sounds colleagues and I weren’t 100% sure we’d be able to hear any of Scott’s phonautograms beyond “Au Clair de la Lune” and the Vocal Scale.

The method our initiative pioneered last year depends on the ability of our partners at Lawrence Berkeley National Laboratory to track waveform images with what they call a “virtual stylus.”  For every point along the time axis, their algorithm needs to figure out where the trace falls along the amplitude axis—like this.  But as we looked at the remaining Scott phonautogram scans we hadn’t yet heard, we saw some problems on the horizon.

In some cases, the side of the stylus seemed to have scraped the surface of the sheet as well as the tip.  In this example, you can tell visually where the waveform’s going, but I don’t know that a computer could follow it as easily.  And sometimes the trace loops back on itself or moves in other ways so that a single point along the time axis corresponds to two or more points along the amplitude axis.  The method we’d been using has to pick one point, so distortion like this runs counter to its logic.

But last fall I ran across a freeware program called ImageToSound, available from Stopmotion Software.  ImageToSound is designed to convert any 24-bit bitmap image into a WAV file as though the bitmap were an optical film sound track. 

As you may know, optical film sound tracks represent changes in amplitude by varying the amount of light that passes through the film—varying either the density or the width.  Those of you who were on the Culpeper tour on Wednesday may remember seeing an experimental negative of a film with an optical sound track on display.  ImageToSound takes digital images in either of these formats and converts them into equivalent WAV files.

By opening a phonautogram in Photoshop and filling in the top and bottom of the image separately, I realized, we can turn the wavy line into two bands of varying width.  Running those through ImageToSound then gives us two WAV files corresponding to the top and bottom of the trace—much like a stereo transfer of a mono recording.  When I make a stereo file from those and reduce it to mono, I get something pretty close to the original waveform shape.  Unless I mention otherwise, this is how all the new phonautograms I’ll be playing during this presentation were converted into sound, although let me assure you the other method is alive and well and currently being pursued further at Berkeley.

One arguable advantage of this approach is that it has a consistent logic to apply to distorted passages in which the trace sometimes moves backwards relative to the direction of recording.  The approach I’m describing doesn’t correct those errors, but because we’re measuring average luminance rather than trying to follow anything, loopbacks and other problems essentially get averaged out, smoothed over. 

We get a playable waveform that may be significantly removed from the sound that was originally recorded, but that at least embodies some basic parameters of the marks on the sheet, such as periodicity.  Phonautograms that had at first seemed not worth trying to play now struck me as very much within reach.

Here’s a Scott phonautogram I applied this method to back in February—undated, but part of the same group as “Au Clair de la Lune” and the vocal scale, deposited at the Academy of Sciences in 1861.  This time the inscription identified the subject matter as speech instead of singing—the opening lines of an Italian play, Aminta by Torquato Tasso—which I thought might just turn out to be the world’s oldest intelligible recording of human speech.  It took a lot of manual cleanup work to isolate the trace made by just the tip of the stylus: here at the bottom you see a half-finished example.  Next we reconstruct breaks in the line, like this, so we can use the paintbucket tool in Photoshop to fill in the top and bottom of the image.

Running those images through the software, and splicing the pieces together, I got this.  What you just heard was played back from the phonautogram at a constant linear speed of 18.375 inches per second, the result of combining a standard CD sampling rate with a scanning resolution of 2400 dots per inch.  But phonautograms weren’t recorded at a constant speed; the cylinder was turned by hand, so the speed varied and wobbled. 

So now it’s time for speed correction.  Here’s the voice track, which we just heard.  Next to it, Scott has recorded the vibrations of a tuning fork.  So we play that back too, and combine the synchronized voice and tuning fork tracks into a stereo file.  Now it’s just a matter of adjusting the tuning fork track to even out the speed fluctuations, and then to set its playback speed to match the frequency of the tuning fork Scott was using, adjusting the speed of the voice track at the same time, just as First Sounds did with “Au Clair de la Lune” and the Vocal Scale.  So here’s the “Aminta” recording set to the same speed as in those other two cases, with the tuning fork at 500 Hz. 

My first thought was that Scott must have recorded this phonautogram at a different speed than the others.  That might make sense—use one speed for speech, the other for singing.  But then I revisited the line of thought we’d followed when settling on a playback speed for the other phonautograms.

Scott had written very clearly on “Au Clair de la Lune” and the Vocal Scale that his tuning fork vibrated at “500 simple vibrations per second.”  So last March, we had to decide what that meant: five hundred of what per second?

After all, what is “one vibration”?  It might mean one cycle: when we measure frequency in Hertz, that’s cycles per second.  But nineteenth century French acousticians measured things differently.  For them, a simple vibration was half a cycle.  What we call one cycle, they thought of as two simple vibrations, or a “double vibration.” 

But Scott makes clear in the writings and diagrams he submitted along with these phonautograms that he didn’t understand vibrations that way either.  What you see on the screen right now is one of his own explanatory diagrams.  From a to b, he writes, is one simple vibration.  From a to alpha is one double vibration.

So what did Scott mean by “500 simple vibrations per second?,” we asked ourselves.  Did he mean 500 of these, or 125 Hz, as his own notes imply?  Did he mean 500 of these, or 250 Hz, as acousticians of the day would have understood?  Or was he confused enough to mean 500 of these, 500 Hz, the unit that seems most obvious to us today?  When we listened to “Au Clair de la Lune” at each of these speeds fourteen months ago, 500 Hz was the only one that seemed close to being right to us.  So that’s what we went with, and that’s what the world heard last March.  And it sounded convincingly like the voice of a young girl.

But this new recording and the other two all displayed voices with a similar frequency range—and when played back to back, they all sounded somewhat similar too—but the new one sounded very wrong.

So when we thought we were hearing the voice of a young girl last March, my colleagues and I now believe, we were really hearing a chipmunk effect—the sound of a much deeper voice, singing very slowly, played back at twice the speed at which it was recorded. 

In retrospect, this would make perfect sense.  Instead of 500 Hz, Scott’s tuning fork would have been vibrating at 250 Hz.  And if any acoustician had provided him with a “500 simple vibration” tuning fork—which is what he wrote he was using—that’s exactly what he would have got—even if he himself didn’t understand what it meant. 

So when you hear these recordings later in this presentation, at the proper points in Scott’s story, you’ll hear them at what we now believe to be the correct speed. 

The concept of the phonautograph starts here, with Longet’s Treatise on Physiology,the book Édouard-Léon Scott de Martinville was proofreading in 1852 when he had his eureka moment.

The mere idea of making vibrations record themselves as wavy lines wasn’t new.  Here’s one example Scott himself cites in his earliest writings on sound recording: in a physics textbook of 1850, Pouillet describes attaching a stylus to a tuning fork, holding the stylus up to the wheel of an electromotor while the fork vibrates, and getting an inscription like this.  He even lists frequencies corresponding to all the notes in a scale he supposedly calculated in this way. 

What was new was Scott’s plan to replace the tuning fork with an artificial human ear that would record sound vibrations from the air, without having to be in direct physical contact with their source.  In looking to nature for design ideas, he was following in the footsteps of his grandfather, who back in the 1780s had modeled his plan for a dirigible balloon on the prototype of the fish.

The practice of phonautographic recording starts here, with Scott’s first actual experiment of late 1853 or early 1854—he later cites both years.  These are paper prints of two glass plates he recorded using paper membranes, one of the sound of a guitar, one of speech.  In print, Scott describes this experiment as taking place with a few of his friends at the house of a musician—without telling us specifically who they were.  But his manuscript copy of the same text has the actual names of the participants, crossed out but still legible.  The musician was Adolphe Giacomelli, also known as a theatrical agent and journal editor, and the friends were a man named Sabbatier and a father and son both named Reboux.

The guitar plate probably came first.  The overall layout suggests that Scott and his friends first tried to move the glass by hand in a spiral or circle, but found this too difficult and switched to recording in straight lines. It wouldn’t have made much sense to record only halfway across the plate unless the other half was already full.

The speech plate would then have come second, after Scott and his friends had switched to recording in straight lines.  If my hypothesis is correct, then the sound of a guitar was the first subject ever recorded from the air using a membrane and a stylus, and the musician Adolphe Giacomelli was probably the person playing it.

What do these records sound like if we try to play them?  Not like much, I’m afraid—here’s one of the more promising traces from the speech plate. 

Scott then wrote up two separate accounts of his idea in early 1857: one in January for deposit in a sealed packet with the Academy of Sciences, one in March as a brevet for deposit with the French patent office—Scott’s fee, in that case, was covered by the Société d’Encouragement pour l’Industrie Nationale, or Society of Encouragement for National Industry, the SEIN for short, which had taken an interest in Scott’s work.  In these documents, Scott laid out both short-term expectations and long-term aspirations.  Eventually, he suggests, his invention might be used to record singing, to preserve the voices of the dead for future generations, and to serve as an automatic stenographer.  But at this point, he was most optimistic that the phonautograph could be used as a tuning device: you’d sing or play a note on an instrument into it, run a plate underneath, and examine the resulting inscription to see if you were in tune or not.  This was one of the few applications Scott could try to develop as long as the flat plate format limited him to brief snippets of sound.

Consider a phonautogram Scott made in March 1857 of notes played on a cornet.  He wrote a different note of a half-tone scale next to each line, illustrating how this method could be used to tune instruments if the recording speed could ever be properly regulated.


Here’s another phonautogram from the same period, this one labeled “phonautography of the human voice at a distance.”  Let’s listen to two complete traces. 

Scott was probably recording single sung notes, and the change in pitch you hear just reflects the recording surface picking up momentum.  The result certainly isn’t recognizable as a voice when we play it, but Scott is picking up a pitch of some kind—which, right now, is all he’s trying to do.

Here’s the “artificial ear” part of the phonautograph as Scott depicts it in his March 1857 patent.  Let’s compare what Scott has in mind to a real ear.  The opening of the funnel corresponds to the external part.  Two membranes correspond to the tympanum or eardrum, and the oval window.  The area separating the two membranes is designed to be adjustable—Scott wants to be able to vary the air pressure, which his physiological readings have suggested is important.  However, no evidence has turned up that Scott actually used anything more complicated than a single membrane to make recordings during 1857.

By July 1857, Scott was using a new format: a paper sheet wrapped around a cylinder.  Instead of making a single quick pass across a plate, he could now record a long, continuous helix.  That month Scott recorded one of thirty phonautograms in this format, dating from the second half of 1857, preserved today in the SEIN archives; Melissa Van Drie and I made a complete set of scans in April.  Now it has become possible for Scott to conceive of recording not just single notes, but, as he writes, a song of the voice—changes of tone—a huge step forward in terms of subject matterThere are twenty rotations here of two traces each—in 1857, Scott often went back after finishing a phonautogram, shifted the cylinder, and recorded a second time in between the traces made the first time around, using the space he later devoted to recording a tuning fork.  Although this is a little anachronistic, you might think of the phonautogram as a two-track medium that can hold one recording in stereo or two recordings in mono.  Scott wrote years later that his phonautograph ran at, quote, “a nearly uniform speed which the hand acquires quickly through practice.”  But when we listen to his results, we quickly learn that the speed was nothing like nearly uniform, and without a tuning fork trace there seems to be no absolutely reliable way to correct the speed.  I’ve settled on 11.5 inches per second constant linear speed for the examples I’m going to play you today.  Here’s what one rotation of the July 1857 phonautogram sounds like at that speed.

Scott seems to have made his first official deposit of four phonautograms with the SEIN with a cover letter written sometime in October 1857.  He writes: “plate 1 was written by an ashen pipe one centimeter in thickness; plates 2, 3 and 4 show the motion of a membrane under the influence of the voice.”

Plate two is signed and dated October 1857 next to the label in ink, but the inscription scratched into the lampblack itself gives the date August 17, 1857—making this the earliest phonautogram we can pinpoint to a specific day.  The subject is “song at a distance,” and there are two enigmatic notations at the beginning and end of the trace: jeune jouvencelle, or “young little girl” at the start, and les échos or des échos at the end – either “the echoes” or “of the echoes.”  Scott could be indicating that he recorded a young girl and some echoes, or these may be lyrics or the names of songs.  Any melody is obscured beyond recognition by the irregular recording speed, so it’s hard to know.  But we do hear hints of articulations—not enough to make out any words, but enough to be able to tell that there are words.

On October 28, Scott gave a lecture at the SEIN which he later had printed for private circulation under the title “Graphic Fixation of the Voice.”

He prepared nine numbered plates as illustrations.  Plate one consists of notes held while singing in a medium-pitched voice. He’s drawn a dotted line to show a waveform he associates with a “husky” voice, and on the right is an example of what Scott calls a wave of inflection, something that excited him because he thought he’d confirmed a theory of the physiologist Johann Müller.

Plate eight, a recitation of the complete Lord’s Prayer in French, gives an idea of what some of Scott’s less successful phonautograms look like.

On November 16, Scott submitted a phonautogram to the SEIN containing an excerpt from the play Othello by Ducis.  Here’s how the passage runs in English translation:  So to this faithless rival Hédelmone must have given this diadem!  In their cruel rage, our lions of the desert, beneath their burning lair, sometimes tear apart the trembling traveler—It would be better for him for their devouring hunger to scatter the scraps of his palpitating flesh than to fall alive into my terrible hands!   This seems to have been Scott’s equivalent of ‘Mary Had a Little Lamb’; he used it as an example again and again over the course of his career.  He included a transcription of the same text, specially marked up, to illustrate his belief that phonautography could bring such declamatory writings to life.  Here are the first four rotations of the actual phonautogram.

Even despite the distortion, we can match the sounds in the recording to the words of the recitation.  S’il faut; I think that’s reasonably recognizable now – qu’à ce rival, a bit less recognizable – Hédelmone infidèle; that’s the right rhythm, at least – Ait remis – I suspect I cut off the end of the phrase; this is all I had time to do before this presentation.  So we have something here that sounds a little like what Scott indicates he recorded—though not much.

On December 9, Scott turned over some more phonautograms supposed to shed light on questions of timbre.  Consider one of two phonautograms labeled “Timbre of the Cornet.”   Let’s hear just the first six rotations of one of the two interlinear recordings on this sheet.  I doubt we’re getting much if any of the original pitches or pitch changes here, but we are hearing something like the timbre of a cornet recovered from the year 1857.

The SEIN had assigned Scott’s case for evaluation to Jules Antoine Lissajous, known for recently pioneering his own method for visualizing sound vibrations, the Lissajous curve, and Lissajous submitted a mixed report from the Committee on Economic Arts on January 6, 1858, commending Scott’s basic idea but observing that the funnel, tube, and membrane all modified the sounds being recorded and rejecting Scott’s claims to have recorded the articulations of speech as the work of an overly vivid imagination.  It’s unclear how Scott took this verdict, but we find no trace of any phonautographic work on his part for the next thirteen months.

In February 1859, Scott was approached by a builder of precision acoustic instruments named Rudolph Koenig, who offered to work on improving his invention.  Scott accepted and signed a contract on April 30, 1859, naming Koenig the exclusive manufacturer and seller of phonautographs.

Until now, historians have generally assumed that the use of a cylinder instead of a plate as a recording surface, as described and illustrated in a certificate of addition Scott filed with the patent office on July 29, was Koenig’s idea.  We now know that isn’t the case.  However, Scott does credit Koenig with giving him the idea of recording a tuning fork time code alongside the inscription of the sound being studied.  Koenig in turn gave credit for this idea to Guillaume Wertheim, who had found in the 1840s that he could measure units of time even shorter than a simple vibration of the tuning fork, since he could divide the wavy trace visually into even smaller parts. 

Following Wertheim’s approach, if you wanted to measure thousandths of a second, you only needed your tuning fork to vibrate at 500 simple vibrations per second, since it was easy to divide each simple vibration into halves, like this.  As I explained earlier, Scott ended up thinking these smaller units were the “vibrations.”


This plate from Scott’s 1859 certificate of addition is the only actual recording we have from this stage of his work, and it was made not with a membrane picking up sounds from the air, but with a tuning fork and a chronometer.  Here’s a simplified illustration of the same experiment from a physics textbook published a few years later: the vibrations of a tuning fork are recorded alongside marks made by a chronometer, so that by measuring the number of vibrations between the two chronometer marks and dividing by the time interval you can calculate the frequency of the tuning fork.  Scott counted 2,613 cycles.  We can imagine Koenig telling him: “Don’t forget, Édouard-Léon, you’ve got to double the number of these double vibrations to get the value in simple vibrations, which are double the units you count to get thousandths of a second when you’re using that other tuning fork I made for you.”  It’s easy to see how Scott could have become confused. 

Koenig went on manufacturing phonautographs for years to come, but Scott soon broke off this relationship, unhappy with decisions Koenig was making about the design of the commercial machine and ultimately blaming him for keeping it in a rudimentary state.

By early 1860, we find Scott working instead with the eminent physicist, chemist, and photographer Henri Victor Regnault. The earliest document pertaining to the phonautograph in Regnault’s papers at the archives of the Institut de France is dated March 28, 1860.  Scott has cut three pieces out of larger phonautograms, each of which he’s labeled as showing the fundamental and harmonics of the voice.  He’s begun recording a tuning fork at 500 simple vibrations per second alongside the voice track, although things don’t seem to be lining up quite right, suggesting Scott was just now figuring out how to do this himself.

Twelve days later, on April 9, 1860, Scott famously recorded “Au Clair de la Lune.”   His technical notes suggest that he understood this as a study of pitch. 

We can reconstruct some details of the machines Scott was using from the dimensions of the phonautograms made on them.  On the right, you see the dimensions of a standard commercial phonautograph cylinder of the 1860s, which was almost a centimeter wider in diameter than the cylinder Scott had been using himself in 1857, as shown on the left.  But the phonautograms Scott recorded in 1860—including the “Au Clair de la Lune” of April 9—were made on acylinder almost a centimeter narrower in diameter than the one he’d been using in 1857.

The distance between traces made by successive rotations on the phonautograms Scott made in 1857 was approximately 1.12 centimeters, and this is also true of the phonautograms he made in 1860.  For comparison, Allen Koenigsberg kindly sent me this scan of the screw on his phonautograph, which turns out to have a double helix that advances just a hair under one centimeter per rotation.  Overall, the phonautograph Scott was using in 1860 didn’t have the same dimensions as known commercial phonautographs of the period—he seems likely to have been using custom equipment.

The Regnault papers in the archives of the Institut de France contain eight complete phonautograms Scott prepared a little over a week after April 9, all within the space of a few days.   On April 17, he recorded another phonautogram of “Au Clair de la Lune,” and another rendition of that passage from Ot[h]ello about ravenous lions.  Judging visually, it looks like he recorded pitch pretty successfully both times, but with only the tiniest hint of timbre, something that’s also true of four phonautograms dated April 18.  Scott’s membrane seems to be picking up the fundamental really well, but not much in the way of overtones.

That changes with a phonautogram dated April 19, when Scott writes that he’s now using “a membrane coated by the process indicated by Monsieur Professor Regnault.”  The waveforms suddenly become much more complex, to the point that the stylus sometimes starts looping backwards.  Scott has coated his membrane in some new way, at Regnault’s suggestion, and suddenly it seems to have become much more responsive to overtones.

.Scott also mentions Regnault and specially prepared membranes in the notes on two of the phonautograms he deposited with the Academy of Sciences in 1861.  This is one of them. Scott writes that this is a “study of tonic accent at the request of Monsieur Professor Regnault,” and his technical notes indicate it was made with three membranes soldered together to cancel out their respective resonant frequencies so that these wouldn’t interfere with the sound being recorded—likely the special coating process Regnault had suggested.  Certainly we’re still seeing those unusually complex waveforms.

Here’s the text, as Scott has written it at the bottom of the sheet, the opening lines of the Italian pastoral drama Aminta by Torquato Tasso: Who would believe that under human form and under this pastoral garb would be found a God?  Not only....(and then it cuts off).  Here’s that phonautogram again, voice + tuning fork, without speed correction, to refresh your memory.  When we correct the speed so that the tuning fork is at 250 Hz, the speech becomes even easier to follow.  As with exhibitions of the tinfoil phonograph, you wouldn’t be able to understand this if you didn’t know what it was in advance—even if you spoke Italian; I’ve tested that.  But we can verify that these are the recorded words, and even that the speaker got a couple words further along in the text than Scott wrote.

Scott has also added an important footnote.  “I was wrong,” he writes, “it should be ‘umane forme.’”  The order of two of the words in the recording is messed up: the official text of the play has “umane forme,” but the recording has “forme umane.”  Scott takes responsibility for that mistake—and in doing so, indirectly identifies himself as the speaker.

So let’s listen once more to the recording, this time just the voice, without the tuning fork.  I give you the voice of the inventor of the phonautograph himself, Édouard-Léon Scott de Martinville, recorded in the year 1860 on his own invention.  If Scott recorded his own voice on one occasion, it’s likely that he did the same thing on other occasions.  Here’s another phonautogram made with a Regnault membrane: the Vocal Scale of May 17, 1860, played at what we now believe to be the correct speed.  That could certainly be the same voice.

And how about the famous “Au Clair de la Lune” of April 9, 1860?  Now, keep in mind that this was recorded before Scott started using the Regnault membrane.  It has more complex waveforms than we see in Scott’s other phonautograms from that phase, but, still, it doesn’t yet reflect the technical advance we heard in the other two examples.  Here’s what it sounds like with the tuning fork at 250 Hz. 

The singing seems painfully slow.  But Scott was aiming to make a record for visual analysis of pitch; this gave him nice, long, regular waveforms to study.  And when I imitated this rendition on a recent visit to Paris, the immediate reaction I got was: ah! that’s how we sing “Au Clair de la Lune” as a lullaby!  We may have to give up our romantic notion of Scott recording the voice of his young daughter, but in exchange we may have a record of the way he sang his children to sleep. 

The last phonautogram in the Regnault papers is dated April 30—yet another version of “Au Clair de la Lune.”  This marks another important shift in Scott’s approach.  He writes that he’s using a tympanum with “natural inclination”—meaning at an angle, as in the ear, something his writings show he’s come to associate with shielding it from stray air currents.  But most interesting of all is the reference to an oval window made of rubber.  This is the first time we ever see notes on a phonautogram that explicitly mention using a second membrane. Scott has clearly returned to trying to imitate the human ear itself.

Four months later, on September 1, 1860, we find evidence of an even more decisive step in that direction.  The subject of this phonautogram is a vocal solo, “Et Incarnatus Est” from Cherubini’s Missa Solemnis in D minor.  Scott writes that it was inscribed by a solid—the footplate of the stirrup—forming a node in the oval window.  The stirrup (or stapes) is one of the ossicles or bones of the middle ear.  Scott has inserted an artificial chain of ossicles made of copper, ivory, and other materials, into his signal chain, with strings pulling on it so that it tugs on the membranes at either end.  Ever since Scott had started work on the phonautograph, he’d believed air was the best conductor of sound, but now he’s changed his mind.

In fact, he’s so enthusiastic about the results he’s getting with his new approach that he runs out a week later and gets two physicists to sign and date it—Professor Gérardin of the Collège Stanislas and Émile Saigey.  He believes he’s finally solved the problem of recording noises—rustlings, consonant sounds, and so forth—both by amplifying them and by attenuating more “musical” sounds.

Scott mentions two collaborators he worked with in this period, both of Russian origin, although I don’t know of any connection with Julius Block: Prince Shahovskoi, who helped him make the chain of ossicles, and Nikolai Khanykov.  Nikolai Khanykov was not a physicist, but an ethnographer who published a well-known Memoir on the Ethnography of Persia a few years later.  In it, he writes at length about the use of photography as a means of gathering data on foreign cultures—something he himself helped to pioneer.  Immediately afterwards, he writes: “The philological instrument is much more difficult to find.”  Alas, he writes, to gather useful linguistic data, you need to be an expert; there’s no way to record speech objectively.  So in September 1860, Scott had apparently found someone who hoped to use his invention as a philological instrument, to make ethnographic recordings.

Next, Scott tried removing the second membrane and attaching his stylus directly to the chain of ivory pieces sticking out towards the cylinder.  

He recorded one phonautogram that way on September 15, 1860.  The subject is a song called “Vole, Petite Abeille,” or “Fly Little Bee.” 

But now Scott decided that he needed to amplify the vibrations for recording by adding an amplifying lever to the end of his signal chain.  This is the last design change he’s known to have made in the phonautograph, and there’s only one known phonautogram that incorporates it.  It’s undated, but presumably the latest Scott recording we have. 
Here it is, another rendition of “Vole, Petite Abeille,” or “Fly, Little Bee.”  As you can see, the amplifying lever gave Scott a larger waveform, but in places it caused the stylus to go utterly haywire.  Still, let’s listen to where Scott was in terms of his ability to record sound after making all the changes to his system I’ve just described.  Here’s the voice track before speed correction.  Now, the same with the tuning fork track added—for technical reasons, I had to take a different approach to playing back the tuning fork that makes it sound a lot like the buzzing of a bee.  Now, the voice track and tuning fork track after speed correction.   What we have this time is not a slow incantation of a scale or song like “Au Clair de la Lune,” but a rendition with some real exuberance in it, as much performance as experiment.  I haven’t been able to identify the song, except that the closing words sound a lot like “Vole, Petite Abeille.”  So it’s possible this recording is the only documentation of the song that exists.  Even if not, it’s an admirable parting gift from Édouard-Léon Scott de Martinville.  Now, speed corrected without the tuning fork.

[Omitted closing paragraph:] The process of recovering phonautographic sounds is ongoing, and I encourage you to follow the progress of the project through our collaborative website, FirstSounds-dot-org.  What’s now the world’s oldest recognizably recorded sound that we can listen to?  I’m not sure—it depends largely on whether you personally think you recognize something in the 1857 phonautograms or not.  More certain is the enduring importance of Scott’s own legacy.  In the end, it’s clear he bequeathed us not just the principle of recording sound from the air using a membrane, but also a rich corpus of primeval recorded sound itself, which in retrospect may be an even more remarkable contribution.



Original content copyright © 2009, Patrick Feaster.