Berthold Laufer’s Chinese Stereo Recordings of 1901

In 1901, Berthold Laufer recorded Chinese music and drama using two phonographs at the same time, one for vocals and one for instruments.  Today we can combine the resulting pairs of recordings in stereo, and I’ll be sharing a few examples below, together with an explanation of how I prepared them.  Laufer’s pairs of simultaneously recorded cylinders predate the oldest previously confirmed stereophonic recordings by over a quarter century.  They may be the oldest true stereo recordings in the world, as well as the only ones from the acoustic era, before the advent of electrical recording.

The cylinders in question belong to a larger collection Laufer recorded as part of an expedition he led to China on behalf of the American Museum of Natural History (AMNH), reporting to Franz Boas.  They weren’t the first field recordings ever made in China—among other things, the cylinders of Chinese dialect readings produced by Paul Georg von Möllendorff are a little older—but their audio documentation of Chinese opera and folk music is still extraordinarily early.  They’re well-recorded from a technical standpoint, and the collection as a whole is also impressive for its sheer size at 503 cylinders total, although these have been split into two separate groups held by different institutions.  Soon after the cylinders had been deposited at the AMNH, 103 of them were sent to the Berlin Phonogramm-Archiv, where they remain today.  The remaining 400 cylinders are now held at the Indiana University Archives of Traditional Music (ATM).  This latter group was recently digitized by Indiana University’s Media Digitization and Preservation Initiative, and it’s currently available for streaming online through the university’s Media Collections Online—albeit only in mono.

Laufer’s correspondence with Franz Boas, preserved at the AMNH, provides important details about his recording activities.  On September 19, 1901, shortly after Laufer had made a few initial recordings in Shanghai of shorter pieces of music, he wrote: “I am going to make arrangements with a band of actors to have them sing one or two complete dramas into the machine and besides to obtain the whole score of the orchestra.”  Then, on September 27, he reported that he’d carried out his plan during the previous week:

I engaged a band of female actors and took the plays on the stage of their theatre.  I used two machines which were working at the same time, one for the orchestral music, the other one for the vocal music, so that two cylinders are corresponding to each other.  The songs bear the numbers 28-52, the corresponding instrumental music = 28a-52a.  The orchestra consisted of 4 instruments, a guitar, fiddle, a drum and castanets.  I had a double funnel, so that all instruments could easily go in, two from each side.

The purpose behind Laufer’s unusual arrangement wasn’t to record performances for playback in stereo.  Instead, he planned to transcribe the texts and the vocal and instrumental music for print publication, which made it useful to have the voices and instruments foregrounded on separate cylinders.  Other early twentieth-century field recordists likewise tried to capture the different parts of complex performances on different cylinders as an aid to transcription, particularly with vocal quartets—see Natalie Curtis Burlin’s Hampton Quartet recordings, for example, or some of Milton Metfessel’s work in “phonophotography”—except that in those other cases, the cylinders of the different parts were recorded one at a time, during separate performances, and not simultaneously.

Let’s now listen to a stereo reconstruction based on one of the pairs of simultaneous recordings Laufer made during the week of September 27, 1901.  The source recordings for this and later examples, presented here courtesy of the American Museum of Natural History’s Division of Anthropology, are held by the Indiana University Archives of Traditional Music and were digitized by Indiana University’s Media Digitization and Preservation Initiative with partial funding from the National Endowment for the Humanities.  Although I debuted these examples at the international workshop “Safeguarding Strategies of Sound Archives in the Pacific Region” at the Shanghai Conservatory of Music on November 5, 2019, this is the first time they’ve been made available for general listening.

遊龍戲鳳 = Yu lung hsi fêng (Wandering Dragon Plays with Phoenix), Part One of Twelve, 41-41a (scy 2768-2769)


The instrumental phonograph starts first, and the vocal phonograph starts a moment later—indeed, you can hear it coming up to speed (a detail which could be corrected, but which I’ve left alone here because I think it’s interesting for purposes of illustration).  At the end, when the cylinder of the vocal phonograph runs out in mid-performance, there’s a shout in the background of the instrumental cylinder—probably Laufer signaling the musicians to stop.  We can plainly hear the sounds of voices and instruments coming from different directions, left and right, which is remarkable for audio from the year 1901, and without parallel outside of Laufer’s work, as far as I’m aware.  However, there isn’t much sound in the center, shared by both channels, to create a “soundstage,” so stereo imaging is limited.  The vocal cylinder contains only a faint echo of gongs and percussion, while the instrumental cylinder contains only a faint echo of voices.  The listener can readily identify two positions—the place where the vocalists were and the place where the instrumentalists were—but there’s not much nuanced representation of relative positioning of performers beyond that.

That said, Laufer went on to make some further recordings that are even more promising from a stereo standpoint, as he reported in a letter dated November 21, 1901:

These records [53-132, 53a-132a] were taken by me this week on Nov. 18/19 with two machines in the same manner as described [previously].  I had a full orchestra consisting of 5-6 men and 4-5 male singers from a theatre.

Let’s next listen to a stereo reconstruction based on one of the pairs of recordings Laufer made with this other group of performers, including a larger orchestra than before.

大香山 = Ta hsiang shan (Great Fragrant Mountain), Part Ten of Ten, 72-72a (scy 2829-2830)


We can hear that the vocalists are still crowded around one phonograph, and that some of the instrumentalists are still positioned by the other one, perhaps playing into a “double funnel” as before.  But it seems that five to six instrumentalists couldn’t all fit closely around the machine—maybe there just wasn’t room—so that the orchestra ended up more widely spread out, resulting in the sounds of the instruments being picked up by both phonographs in more varied proportions.  There had been such cross-over before—after all, there had to be some in order for Laufer to coordinate the parts in his planned transcription—but now there’s significantly more of it, to the point that the listener can form a decent mental picture of how the performers were arranged in space, not just in two clumps, but in a kind of arc.  The illusion of multi-directional auditory perspective becomes more compelling, more immersive.  It’s still a far cry from what we’d expect of a competently-engineered stereo recording made today, or even in the 1930s.  But for the 1900s, it’s stunningly unique.

Laufer also reported making some other recordings in his letter of November 21: “Besides the dramas I have taken a number of popular songs in the Shanghai dialect and pieces of instrumental music as duets on flutes and trumpets (130-132a).”  For our third and final example, let’s listen to a stereo reconstruction of one of the “popular song” segments.

Shanghainese popular song, possibly Part Three of Five, 127-127a (scy 2921-2922)



Stereo sound has a longer history than you might think.  The word stereophonic is often said to have been coined by Western Electric in 1928, but in fact Alexander Graham Bell had already used it back in 1879.  Some nineteenth-century transmissions of live performances by telephone were stereophonic, including those of the Théâtrophone, through which customers could listen to opera with a separate receiver at each ear.  Harvey Fletcher and Alan Blumlein are widely credited with having made the first intentional stereophonic recordings in the 1930s, but Franklin M. Doolittle had been working on the idea since at least 1921 (see patents here and here); whether he really made any such recordings, I’m not sure, but he did pioneer actual stereophonic radio through his station WPAJ in 1924.  And even before that, the “Multiple Graphophone” patented by T. H. Macdonald—filing date June 11, 1898, issue date October 21, 1902—had been equipped to make simultaneous multi-track recordings of “concerted music” that seem to qualify in retrospect as stereophonic.  An article in the Phonoscope for March 1898 describing a similar device—but credited there to Leon Douglass—explained:

If a quartet is going to place a song on the cylinder four horns on as many recorders are placed in position.  One horn will be placed at the left extremity of the cylinder, a second one one-fourth of the length of the cylinder, farther to the right, a third one in the middle and the fourth one three-quarters of the length of the cylinder from the left end.  Each horn will then have before it a fourth of the length of the cylinder to travel over toward the right before it will interfere with a record made by another horn.  The machine is started and each member of the quartet sings into a horn of his own and a perfect and distinct record of his voice is made.  Of course, when the reproduction is made four horns on four reproducers are used.  It has been found by experiment that the result is incomparably superior to that effected when only one record is made by all four voices.  By using the separate records a clearness and distinctness of each voice is obtained, much like that heard when the voices themselves fall on the ear.  This was impossible with the single record.

T. H. Macdonald, “Multiple Graphophone,” U. S. Patent 711,706, filed June 11, 1898, issued October 21, 1902.

No such multi-track cylinders from that period are known to survive, but they apparently once existedotherwise nothing could have been “found by experiment”and since all the tracks were recorded onto a single cylinder, synchronization in playback would have been easy to achieve.

Meanwhile, the earliest known surviving specimens of stereophonic sound have until now fallen into a category known as “accidental stereo”: cases in which sounds just happened to have been recorded simultaneously from multiple perspectives without anyone specifically intending to capture multiple perspectives, but in such a way that they can nevertheless be combined to produce a stereo effect.  Beginning in the late 1920s, shortly after the widespread mid-decade adoption of electric recording, two different microphones were sometimes used to feed a pair of recording lathes to cut two separate disc masters from a performance simultaneously so that one of them could serve as a backup or alternate version.  This happened relatively rarely—ordinarily a single microphone fed both lathes—but when it did, and when both masters can be identified, it opens up the possibility of recreating “accidental stereo” from them.  Experiments along these lines date back many years now; see, for example, Barry Fox’s article “Mono sounds reveal stereo secrets” in New Scientist (December 1985), 59-61, highlighting work by Brad Kay.

The difficulty lies in synchronizing the two masters well enough to yield a decent stereo effect, which turns out to be quite a challenge.  The two masters would have been recorded onto two discs using two separate turntables, each burning its own irregularities in rotational speed into the original.  Meanwhile, the process of digitizing analog discs introduces further speed irregularities of its own.  In addition to variations in the rotational speed of the turntable used for playback and the nontrivial matter of disc-centering, there are also timing issues associated with individual combinations of tonearm, cartridge, and stylus.  For evidence of that last point, we can take any set of transfers made by George Blood LP for the Great 78 Project—distinctive for using four tonearms with different styli to track the same disc simultaneously—and use cross-correlation to determine how the four transfers line up with each other.  Here’s a graph I made of one typical example (Silas Leachman, “Don’t Forget to Write Me Every Day,” Victor Monarch 1122), with the vertical scale in audio samples at 96 kHz:

Not only do we see gradual drift in the timing of individual transfers over the course of the whole disc, but we also see each transfer oscillating around ten samples backwards and forwards with each rotation.  And remember, we’re dealing here with the same disc being played on the same turntable at the same time.  I ran into these problems while trying to average transfers of multiple pressings of the same disc to factor out wear and tear, which would be a valuable technique if it could be made to work, but which is surprisingly difficult to get right.  Doing the same thing with the two halves of a stereo pair would be even more difficult because the sources wouldn’t have been identical in the first place, and also because slight inaccuracies in relative timing could jeopardize the stereo effect, particularly if that effect would have been fairly subtle to begin with.  That said, Celemony’s Capstan—a piece of software that can mitigate wow and flutter—has been brought to bear on the “accidental stereo” problem with gratifying results: see the Pristine Classical release Accidental Stereo, 1929-1933 (PASC 422), representing work by Mark Obert-Thorn and Andrew Rose.

Before the mid-1920s, sound recordings were as a rule made acoustically rather than electrically—that is, with cutting equipment powered by the sheer mechanical force of the sound waves themselves.  There doesn’t seem to have been much room for “accidental stereo” in the commercial recording industry of the early 1920s, or during the decade or two before that either, since the onerous studio practices needed to create one perfect master recording for mass duplication in that period wouldn’t have permitted cutting simultaneous backup copies.  During the 1890s, on the other hand, cylinder recording companies had almost always captured each performance on several phonographs at once, the goal being to cut more records at a time when the number of duplicates that could be made for sale from any one original was relatively limited, and when originals were also sought after for their significantly higher sound quality.  Any two cylinders cut during the same “round” would together have comprised a stereo pair in exactly the same sense as the discs cut on two lathes fed by two different microphones in the late 1920s, although their vantage points might have been displaced vertically rather than horizontally.  The Liebig chromo shown below actually dates from the first decade of the twentieth century, but it provides one of the most attractive illustrations I’ve seen of a typical 1890s studio arrangement.

So if nearly all commercial cylinder recordings of the 1890s were made under circumstances like these, why has reconstructed 1890s stereo proven to be so elusive?  There are a few reasons.  First, the survival rate for brown wax cylinders is poor enough for the odds of finding two from the same “round” to be rather slim.  Second, different studio arrangements would have been more or less conducive to stereo imaging: recordings of bands and orchestras might be promising, but the arrangement in the chromo shown above would be less so, and a similar arrangement centered on a single soloist even less so than that.  Third, even if we were to find two recordings from the same “round,” made in theoretically optimal positions for stereo imaging, we would still face daunting challenges when it came to synchronizing them with each other.


Stereoscopic anaglyph assembled from two tintypes taken with two vertically separated lenses, from my October 2015 blog post, “Historical 3D Fun With Anaglyphs.”   Pull out your red-cyan glasses!  This obviously isn’t how an intentional stereoview would have been contrived.  With sound recordings, “accidental stereo” might often be comparably ill-starred.

At the 2011 annual conference of the Association for Recorded Sound Collections, John Levin and Dan Reed gave a presentation entitled “1891 Brown Wax Stereo”; you can listen to audio of it here.  The subject of their experiment was two copies of an 1891 Edison recording of “Cujus Animam” by cornetist D. B. Dana—one of them is available here—which they aligned using a cough in the spoken announcement as a sync point.  Their results, as they acknowledged at the time, were inconclusive.  There was some illusion of spatial depth, but it might just as plausibly have come from slight misalignment and differences in equalization—analogous to the simulated stereo of the 1960s—as from true stereo information documented during the recording session itself.  Indeed, the two cylinders they used are likely to be pantographic duplicates cut from the same master (the announcement is identical in both cases, whereas contemporaneous practice at Edison’s laboratory, judging from this letter, seems to have been to record the announcement on each cylinder separately from the main performance; on the word “duplicate” in the announcement, see also this letter).  Even so, this presentation was an important milestone in the history of paleostereophony.  I’m not aware of any other attempt made to reconstruct stereo from a pair of cylinder recordings of the 1890s, so there’s that; but perhaps even more significantly, Levin and Reed spelled out the daunting technical obstacles that stand in the way of such efforts, and which anyone would need to overcome in order to recover stereo from a pair of commercial cylinders made under the “round” system.

Many of the same issues pointed out by Levin and Reed also affect Laufer’s cylinders.  His two phonographs were run by spring motors that were liable gradually to speed up or slow down at different rates and different points, and that also produced lots of rapid flutter.  Below is a graph of the wavering pitch of the gong sounded at the opening of 28a (scy 2747), the instrumental cylinder belonging to Laufer’s first stereo pair, together with an audio file that presents, first, the unaltered original, and then a version with the pitch fluctuations evened out to show how much of a difference their presence makes.


I stabilized this excerpt using some code of my own, described here, just by locking the pitch of the gong to a consistent frequency, but speed stabilization isn’t generally that easy, since there are usually multiple instruments or voices in play.  Existing software—including Capstan—seems ill-equipped to stabilize the speed of most of Laufer’s Chinese recordings.  And as long as the speed of both individual source recordings remains unstable, then the synchronization between them will obviously be unstable too.

But for present purposes that might not matter so much.  If Laufer’s cylinders had been recorded with a pair of horns both aimed at the same subject, as in the commercial “round” system, it might be nearly impossible to synchronize them well enough to produce a stereophonic effect because the differences between them would be subtle and easily scrambled.  Fortunately for us, the differences between Laufer’s paired cylinders are not subtle.  In fact, they differ so much that it can be challenging to find sync points that clearly correspond to each other.  But as long as we can align them closely enough that any sounds audible on both cylinders will seem to occur at the same time, we’ll experience a passable illusion of hearing the performance coming at us from two directions.  In other words, Laufer’s cylinders are much more forgiving as stereophonic source material than commercial cylinders from the “round” system would have been.

Here’s the procedure I used:

  1. Find a pitched sound near the beginning of the performance that’s audible on both cylinders—usually a gong—and measure its frequency.  Then adjust the speed of one of the cylinders so that the frequency of the sound is identical in both.  This first step is necessary because Laufer’s two phonographs weren’t necessarily set to run at the same speed, and also because each digital transfer was made independently at a speed chosen by ear without reference to the other cylinder in the pair.
  2. Find a sync point near the beginning of the performance and use it to align the two recordings as left and right channels in a stereo file.  This step is equivalent to what Levin and Reed did with the cough in the announcement of “Cujus Animam.”
  3. Identify a dozen (or so) more unambiguous sync points, making a list of corresponding sample numbers.
  4. Using a piece of custom software written for this project, plot a smooth curve through the average positions of the sync points and resample the two recordings based on it, such that the sync points in both recordings will line up with no abrupt speed changes between them.
  5. Adjust the playback speed by ear to something that sounds natural.

You’ve already heard some results (presuming you listened to the examples shared above); however, please bear in mind that these aren’t yet full restorations, but only proofs of concept.  I built my stereo reconstructions from access files that have had some quick and basic restoration work done on them by Dan Figurelli and Melissa Widzinski of MDPI, but undertaking a serious restoration would mean revisiting the unaltered preservation files and (among other things) restoring them consistently with each other.  The resampling algorithm for aligning the files could use some more work too, since I think it might be producing some unwanted artifacts.

But in the meantime, I want to argue that such stereo reconstruction is necessary to do justice to this source material, and not just an optional novelty.  It’s true that the paired cylinders weren’t intended for playback together in stereo—which wouldn’t have been technically possible in 1901—but if we care about intentions, we may as well also acknowledge that the audio wasn’t intended for public consumption in the first place.  Laufer made these pairs of recordings so that he could transcribe them for print publication, not so that others could listen to them.  His transcriptions would presumably have “synchronized” the parts in the sense of notating them on parallel staves, and although he never got around to making such transcriptions himself, a student in Germany named Erich Fischer did, as shown below (the staves unlabeled here are linked at the start of the transcription to voice, fiddle, and yangqin).

Not only are the separate cylinders not very compelling to listen to in isolation, containing long passages in which content can be heard only faintly in the background, but they were originally created to function as two parts of a coherent whole and to document a meaningful performance only when taken together.  Stereo reconstruction respects Laufer’s original purpose, and the integrity of the original performances, in a way that monophonic presentation wouldn’t.

So is this a case of “accidental stereo,” because the recordings weren’t meant to sustain a stereophonic illusion in playback, or a case of “intentional stereo,” because the two different perspectives were captured very much on purpose?  You be the judge.

Berthold Laufer Stereo Cylinderography

Inscription on bottom of box for cylinder “3a” (scy 2726).

Sticker on side of box for cylinder “P. R. 800” (scy 2970).

Several different numbering systems have been applied to Laufer’s Chinese cylinders over the years.  The current shelf numbers for the group held at the ATM, assigned in the 1980s, are scy 2726-3125, where “scy” stands for “small cylinder.”  The AMNH had previously assigned a different range of numbers to them sometime before part of the collection was sent to Berlin.  These numbers are found written on stickers attached to the sides of the cylinder boxes, and they have the prefix “P. R.,” which I assume stands for “Phonograph Record.”  They run from 478 to 978, with two cylinders assigned to one number (630a and 630b), probably by mistake, and one broken cylinder unnumbered (scy 3125).  The Berlin Phonogramm-Archiv also assigned “inventory numbers” to its part of the collection.  Finally, there’s Laufer’s own original numbering system, which runs from 1 to 392 (again with no number assigned to scy 3125).  These numbers are written on the bottoms of the cardboard cylinder boxes, sometimes accompanied by other notations.  Laufer often assigned the same number to multiple cylinders, either because he reused numbers by mistake (as seems to have been the case with 223-226 and 223a-226a) or because the cylinders were related to each other in some way (as with 3a-3c for three sequential parts of the same piece).  Laufer assigned each of his stereo cylinder pairs a single number, distinguishing the instrumental cylinder from the corresponding vocal cylinder with the letter “a.”  Thus, in Laufer’s numbering, the first stereo pair consists of items 28 (a vocal cylinder) and 28a (the corresponding instrumental cylinder).  Selection titles are given below as they appear in early documentation, both in their traditional Chinese characters and contemporaneous romanizations.  All of Laufer’s stereo pairs, and most of his other Chinese recordings, belong to multi-part selections that were too long to fit on one cylinder and so had to be spread out across several.  There often seems to have been an effort to stop and re-start the performance at logical breaking points, but if a performance simply ran off the end of a cylinder, Laufer had the performers back up and repeat the last segment with more or less overlap.  I haven’t found any instance in which Laufer seems to have used two phonographs to bridge the gap between cylinders, starting one shortly before stopping the other, even in the case of the recordings he made in Beijing in 1902.  By that time he’d reverted to recording all performers together on single cylinders, but his inventories of equipment list only a single phonograph, so he might have had to borrow a second machine for his sessions in Shanghai and not had it available to use later.

Shanghai, week ending September 27, 1901

  • 桑園會 = Sang yüan hui (Meeting in Mulberry Garden), 28=pr507=scy2746 to 38a=pr528=scy2767 (11 pairs, 22 cylinders total).
  • 翠屏山 = T’sui ping shan (Cuiping Mountain), 39=pr529 to 40a=pr532 (2 pairs, 4 cylinders total), at Berlin Phonogramm-Archiv.
  • 遊龍戲鳳 = Yu lung hsi fêng (Wandering Dragon Plays with Phoenix), 41=pr533=scy2768 to 52a=pr556=scy2791 (12 pairs, 24 cylinders total).

Shanghai, November 18 and 19, 1901

  • 大香山 = Ta hsiang shan (Great Fragrant Mountain), 53=pr557=scy2792 to 72a=pr595=scy2830 (9 pairs plus an unpaired cylinder, comprising 10 segments; 19 cylinders total).  The unpaired cylinder is 61a=pr573=scy2808, instrumental with faint voices audible in the background, with no vocal counterpart 61 and no corresponding gap in the AMNH numbering.  It seems the vocal part for that segment either was accidentally not recorded or went missing at an early date.  Damaged and not transferred as of 2019: 62a=pr575=scy2810, 71=pr592=scy2827.
  • 戰北原 = Chan pei yüan (Battle of Beiyuan), 73=pr596=scy2831 to 82a=pr615=scy2850 (10 pairs, 20 cylinders total).
  • 黑風帕 = Hei fêng p’a (Black Wind-Kerchief), 83=pr616 to 91a=pr632 (9 pairs, 18 cylinders total), at Berlin Phonogramm-Archiv.
  • 黃鶴樓 = Huang hao lou (Yellow Crane Tower), 92=pr633=scy2851 to 96a=pr642=scy2860 (5 pairs, 10 cylinders total).  Damaged and not transferred as of 2019: 95=pr639=scy2857.
  • 五台山 = Wu t’ai shan (Mount Wutai), probably followed by three other dramas, 97=pr643=scy2861 to 124a=pr698=scy2916 (28 pairs, 56 cylinders total).  Damaged and not transferred as of 2019: 105a=pr660=scy2878.  Laufer states in his letter to Boas of November 21, 1901, that the cylinder range 53-132 contains “8 longer dramas” including Ta hsiang shang, but only five are identified by name in the lists I’ve seen.  I suspect that the other three dramas followed Wu t’ai shan in the sequence without being separately identified.
  • 125=pr699=scy2917 to 129a=pr708=scy2926 (5 pairs, 10 cylinders total).  Laufer states in his letter to Boas of November 21, 1901: “Besides the dramas I have taken a number of popular songs in the Shanghai dialect and pieces of instrumental music as duets on flutes and trumpets (130-132a).”  Since cylinder pairs 97-124 all sound like parts of dramas, while pairs 125-129 share an instrumental melody, with singing but no speech, I infer that the latter five pairs must correspond to the “popular songs,” even though they’re not identified as such in lists.

The remaining cylinders labeled as pairs with numbers 130-132, such as 130 and 130a, seem not to have been recorded simultaneously but to contain independently captured instrumental performances—the “pieces of instrumental music as duets on flutes and trumpets.”  I suspect Laufer had labeled a number of pairs of boxes ahead of time in the interests of efficiency and then didn’t relabel the few which he ended up using to make single recordings.  Further support for this view comes from the fact that the boxes for some later cylinders to which Laufer assigned numbers 276-309 also display crossed-out numbers in the ranges 133-152 and 135a-152a, all scrambled out of order.  Thus, Laufer had probably gone into the recording sessions in November with pairs of boxes pre-labeled up through 152 and 152a.

So how much actual stereo content could this list represent?  If we estimate the average duration of Laufer’s cylinders as three minutes apiece—which is probably on the low side—we get the following totals:

  • Total stereo pairs: 91 = ~4.5 hours
  • Stereo pairs at ATM: 80 = ~4 hours, including four pairs in which one of the two cylinders is broken, or badly cracked, and not yet digitized
  • Currently digitized stereo pairs at ATM: 76 = ~3.8 hours
  • Stereo pairs at Berlin Phonogramm-Archiv: 11 = ~33 minutes.  A few of the Laufer Chinese cylinders in Berlin are reportedly broken, but I don’t know whether any of the stereo pairs are affected.

It’s possible to retrieve content from many broken and badly cracked cylinders, although this is a relatively expensive process, so we might eventually be able to assemble stereo versions of everything in the above cylinderography except for 61a (since its vocal counterpart is missing rather than merely damaged).

Broken cylinder scy 2878, or 105a in Laufer’s original numbering: one half of a potential stereo pair—possibly Part Nine of 五台山= Wu t’ai shan—but not yet digitized.


I can think of one other category of sound recording which might arguably count as true stereophony, and which dates back long before Laufer: namely, phonautograph records made with tuning-fork reference traces, including Édouard-Léon Scott de Martinville’s famous April 9, 1860 phonautogram of “Au Clair de la Lune.”  In such cases, the sounds of the voice and tuning fork were both captured simultaneously, and we can play them back together in stereo today, listening to the voice in one channel and the tuning fork in the other.  Both of these sounds would legitimately have been present during the original recording sessions, and sometimes Scott even seems to be singing along with the tuning fork as a drone, suggesting that he was able to hear it and that it influenced his choice of key.  Take his April 20, 1860 recording of “Au Clair de la Lune,” for example:

Of course, Scott wasn’t trying to record a subject from two perspectives, as Laufer was.  He was trying to record a subject from one perspective while using another, simpler sound only as a timing reference.  Whether we accept his phonautograms as the stuff of stereo depends, at least in part, on how we understand the tuning fork: as an integral part of the events Scott was recording or as an extraneous pilot tone.  But I bring this point up here to emphasize that the “firsts” of sound-recording history are hardly ever straightforward, and that the “world’s oldest stereo recording” is no exception.

NB. Edison Papers link updated September 29, 2022.

12 thoughts on “Berthold Laufer’s Chinese Stereo Recordings of 1901

  1. Very interesting! Excellent post! A few years ago I upmixed an 1891 cylinder recording to stereo through the use of sound source separation and spectral editing: Check out the stereo vocal harmonies in this 1904 recording: Those of you interested in upmixing mono recordings to stereo should check out my website, ( Be sure to check out the many examples on the MEDIA pages of the website! I had also started a companion Facebook page for the website ( where I post updates and related content. Be sure to “Like” the Facebook page to follow it for the updates!

  2. Pingback: Berthold Laufer’s Chinese Stereo Recordings of 1901 — Griffonage-Dot-Com – INSSAIDOR

  3. Hi Patrick! Great article! I wish I had known about this when we were doing the transfers. How did you discover that these were recorded in this way? Are Laufer’s field notes present in the ATM collection? Or did you have to track them down elsewhere?

  4. What a coincidence to see this here! I was just going through Laufer’s recordings available on the Indiana University website a month ago. I paired one of his recordings with a restoration I did of an excerpt from Auguste François’s 1902 films in China (just for background music): It just so happened to be one of the selections (SCY 2922) that you presented here! I hope to see more work done on these important cylinders in the future.

  5. Pingback: Do These Tubes Hold the Oldest Known Stereo Recordings? - Atlas Obscura

  6. Pingback: Do These Tubes Hold the Oldest Known Stereo Recordings? - Non Stop Travel

  7. This is a great achievment!, I propose also, to restore some sterephonic tracks, to use the demucs IA techonology to separate the tracks, and then remix as always

  8. Pingback: Using Audio to Map the Surface of a Phonograph Cylinder | Griffonage-Dot-Com

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.