Many thanks, again, for your translation. It was fine to gain the jist of the article, and the
article was quite interesting. This concept holds a lot of promise. I look forward to the day when
there will be a device that is about the size and cost of a large-format document scanner onto
(into) which you place a grooved disk and then a digital conversion takes place, the net result
being a digital sound file. Ideally, scanning and conversion would be some fraction of actual
-- Tom Fine
----- Original Message -----
From: "DECLERCQ Brecht" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Monday, January 10, 2011 6:16 PM
Subject: Re: [ARSCLIST] ELP - Clareety
You'll find my translation of the French article about Clareety beneath (and completely for free!).
I hope there are not too many errors as French nor English is my mother tongue. And in fact I'm
neither a specialist of these disks.
I'd like to draw everybody's attention to the fact that this is an article of October 2007.
I don't know what the status of the commercialization of this device is, but I hope that the fact
that it was again presented in Paris at AES a month ago means it will soon be available.
Media Manager DiVA-project
VRT - Room 2F3
The digital image serving the digitization of old disks
Jean-José Wanegh (1st October 2007)
Translation by Brecht Declercq
As a photography and phonography passionate, Charles Cros wanted to go beyond time. It's curious to
see that today, sound meeting image gives us the possibility to make his dream come true. In this
way, millions of phonographic records, threatened by years of time and victims of the fact that each
play-out wipes out a little more the memory of these supports, are going to find a new youth, thanks
to optical procedures without contact. One of these, Clareety, developed by a research team of INA,
has numerous characteristics responding to the needs of archivists and restoration-specialists.
Although the audio quality of CD's is considered very good by the majority of listeners, there is a
population of audiophiles who consider the encoding at 44.1 kHz of CD's not sufficient for a
trustful reproduction of the musical treasures recorded with analog means. As an answer to this
expectation, there is a tremendous offer of turntables for microgroove records. But even with the
high level performance of these devices and the caring of these audiophiles for their vinyl records,
every play-out still means a kind of wear-out. And although the pressure force of the read head is
adapted to a minimum, there still is a mechanic contact of the needle with the sides of the groove.
With every passing by, this slow scratching leads to an irreparable deterioration of the disk.
That's why as early as 1982, a student of Stanford University, Robert Stoddard, proved the
possibility to use a laser, connected to a special arm, to realize an optical read-out without
contact with the vinyl. In 1983, he created a firm called Finial Technology and with the help of
Robert Stark he developed the first vinyl turntable reading out with laser. In 1989 Finial sold its
patents to ELP from Japan. Only in 1991 the first ELP turntable was commercialized, while already
for two years the CD sales had doubled LP's.
An audio heritage in danger
But the universe of the disks should not only be considered from an audiophiles' view, neither by
the view of a nostalgic in search of lost time, when it was Teppaz to entertain our juvenile
parties. As the years passed by, millions of recordings have been made around the world. For many
years, phonographic recording was the only way to consign a musical, artistic, historic of political
event to a physical support. Thus, an considerable amount of recordings is part of our heritage.
Let's not forget that although it was invented in 1935, the magnetophone tape only appeared at radio
stations after the Second World War.
>From the thirties until half the fifties, radio stations used directly carved disks, called disques
>souples, commonly known in France as Pyral. We should preserve all these recordings, but we should
>also make it possible to listen to them in the best possible conditions, without compromising the
>integrity of the supports. The problem is that we're talking about original recordings with often
>only one known copy. The weight of time made them fragile to the point that even one reading-out,
>even with a performant phono-pick-up, would destroy it forever. This motivated strongly throughout
>the whole world the development of no-contact read-out systems, with a diversity of optical
>solutions, from simply capturing the image of the disk with a scanner, to 3D-capturing of the
>topography of the groove using the principles of confocal laser microscopy. But sound archives'
>leaders don't choose an appropriate solution following the same rules as those followed by an
>audiophile. The reason is that at INA, the collection to be safeguarded counts 276.000 grooved
>disks (fig. 3). Even if optical read-out is the only way to extract the precious content of these
>disks without compromising their eternal life, still this chosen technique should make it possible
>to work without delay, with an optimal quality, using tools that can be industrialized and thus at
>an affordable cost of purchase and exploitation. If one knows that some technologies ask up to ten
>times the real time playing duration of the support for capturing them in 3D, no doubt this may be
>called a handicap that a lot of institutions would not be able to overcome. All these arguments
>were thus considered, when a team of INA researchers completed an original system for the optical
>analysis of the groove of a disk by colorimetric encoding of the variations of the gradient, in
>relation to the axis of the track.
ELP, the first laser turntable serving wealthy audiophiles
In this article, unfortunately it is not possible to make an overview of the range of all the
solutions making use of optical read-out, going from those that are planned for the future, from
whom some already have been commercialized, or at the point of industrialization, to those still in
the prototype stadium. All these systems can be classified in multiple ways.
Firstly, there are systems working in real time, with the optical information directly being
converted in an analog audio signal. In this system there is no signal processing. The ELP turntable
is of course a typical example here. The light produced by a laser is divided in five rays. Two of
them are meant to follow the track of the disk by detecting the 'shoulders' of the groove (the tops
of the sides). After the reflection, these rays are digitized and then treated to serve tracking.
Two other rays lighten the flanks of the groove on an adjustable height, to benefit from a zone
where the needle did not leave traces of wearing the groove out (fig. 6).
The reflected and modulated light of the sides of the groove is sent back to a detector that
converts the light signal into an analog audio signal. The fifth ray allows to control the altitude
of the optical pick-up arm, in function of deformations of the disks and to adapt to different
widths. If the barrier of the cost of these turntables is overcome, this solution may have several
advantages to convince audiophiles and one would be tempted to believe that already with this device
we have the ideal tool for the transfer of all phonographic collections of the whole world. But
nothing is less true in fact. It is an excellent tool for reading out vinyl and 78 rpm LP's in
excellent condition. The majority of tests by audiophiles speak with praise about this technique,
but the systems has huge problems to cope with scratches and dust. The system is unable to read from
transparent or colored vinyl, or materials transparent to the wavelength of the laser. We see the
same impossibility with directly carved acetate disks. It is impossible to read the 78 rpm's with a
vertical groove, as the rays have been conceived to detect the lateral modulation. It's also
important that the form of the needle is V-shaped. Round-carved sides of the groove could create
distortions. To assure a correct following of the track, it is necessary that the top of the needle
and the surface of the disk come together in a perfectly defined angle. In case of a border rounded
off, the tracking will be impossible. The last issue is that scratches or bits of dust could deviate
the system from the groove, in some cases à cheval (spread) over two adjacent tracks, one ray
reading the side of one groove, the other reading the opposite one of its neighbor.
In the nineties Juraj Poliak, an associated researcher of the Ecole Polytechnique Fédérale de
Lausanne (CH), developed a real time read-out system using an optical fiber, with a glass ball on
its edge, to follow the groove of a disk or a cylinder. This fiber allows to take a laser ray into
the groove, that after the reflection towards a photo-detector takes into account the lateral and
vertical shifting of the original cell. This system is very simple in its concept and it has the
advantage of being able to read the lateral grooves as well as the vertical ones. Poliak is now
retired, and his method seems never to have gone beyond the prototype stadium.
Read-out without contact by digital imaging
After these real time systems came the systems that in a first stage produce an image of the disk or
the groove, measuring then its width, to finally extract the digitally processed audio information.
On this terrain, two projects are taking place. One of them, named IRENE, is lead by a team of two
researchers, Vitaliy Fadeyev and Carl Haber of the Lawrence Berkeley National Laboratory. The other
project one was born at the Lausanne School of Engineering, and lead by Ottar Johnsen, with active
participation of Sylvain Stotzer, who made his PhD in information science on this subject. The team
of Berkeley works in the field going from applied optical metrology to particle physics, the latter
being the source of the idea to make a trace of micro-images of the surface of the disk, that after
processing will allow to reconstruct the complete image of the groove. Typically these are 700 µm to
540 µm pictures, containing multiple parts of grooves (fig. 7). In case of a 78 rpm disk of 25 cm,
this means 100.000 snapshots have to be made for a data-volume of 1 Gb to 1 Tb. On this level it is
possible to process the image, to eliminate certain defaults caused by particles, scratches or the
disk being worn-out. After photographing, it's all about measuring in every point the position of
the border of the groove, compared to the middle of it. With this technique, only the disks with a
lateral groove can be treated. Disks with a vertical groove, like cylinders, are analyzed with a
confocal microscopy-device using laser, which is a very heavy technique to setup. Ottar Johnsen and
Sylvain Stotzer's approach is very different. They make one high resolution picture of the disk,
which is digitized afterwards via polar coordinates, with a scanner specially developed for this
purpose named VisualAudio. Here too a precise identification of the borders of the groove has to be
done to measure their position.
Whether it's about this Swiss project or the one of Berkeley, in the two cases the subject of the
captured information is the position of the sides of the groove. But, as many of you know, the
amplitude of the audio signal is not proportional to the position of the needle, but to the speed of
its lateral shifting (fig. 8). And thus, to know this speed, we have to calculate the derivation of
the curve that determines the position of the groove. This makes it necessary to have an excellent
resolution to measure the position of the groove at its beginning, typically 1 micron per pixel.
In France, INA prefers to see a mirror in every disk
In France a research team of INA directed by Jean-Hugues Chenot has chosen an original direction,
starting form a relatively simple but long known constatation, to develop its optical digitization
unit for old disks. The idea is to keep in mind that every side of a groove behaves as a mirror.
It's important always to remember that using a lateral modulation, the radial speed of the grooving
needle is proportional to the amplitude of the signal. So every angle formed by the side of the
groove and the tangent of the circle that describes the rotation of the disk is a result from the
combination of the radial speed of the graver and the tangential speed of the disk (fig. 10 and 11).
If now we lighten the flank of this groove with a bundle of light with certain characteristics, and
we analyze then the reflected light, we should be able to discover the sound information that is
recorded on the disk.
There is a precedent of this method, known by all sound engineers who have engraved disks as the
Buchmann-Meyer method. This method was used to verify the engraving-characteristics of their device.
Rather than measuring with a microscope the amplitude of the groove, one measures the width of the
luminous strip, reflected by the side of the groove when the disk is lit from a distance with a
parallel bundle of light. In every point the width of this strip is proportional to the speed of the
engraving, and thus to the amplitude of the sound. This measuring technique was already used for the
engraving of 78 rpm's. These engraving test disks were called waxographs.
But the difference with the method developed by the research team of INA is that this method is used
only as a way to control things. It doesn't allow to optically read-out the content of the disk.
Reading out the sound by a colored, structured illumination
The great originality of the method proposed by Jean-Hugues Chenot is to encode the incident light
in various colors. So, at the inside of the continuous colored spectrum coming from a filter at the
end of an optical condenser, every colored ray is characterized by his incident angle. While this
rainbow-colored bundle, of whom all rays that converge into one point, hits the side of the groove,
every elementary colored ray is reflected in an angle that is the same as its incidental angle
(always conparing to the norm) in this point of the groove (cf. Descartes' law of reflection). If we
put a sensor in a well-determined angle on the path of the reflected bundle, only the colored ray
with the right incidence (compared to the gradient of the side of the groove) will be reflected in
the direction of the sensor (fig. 12 a, b, c). Doing so, we have an encoding related to the color of
the angle of the groove.
If instead of a simple sensor we put a color video camera, it's not a point per point analysis
anymore. Within the field covered by this camera, multiple parts of the groove can be simultaneously
analyzed. Typically every image covers a zone of 2mm by 2mm and includes eight tracks (in the case
of a 78 rpm), of which six are taken into account with every analysis (fig. 13). We thus obtain a
useful image of 400 by 400 pixels. But one of the big advantages of this method, compared to the
others that have been described earlier, is that this camera carries out a high speed measurement of
the whole disk. From a measurement from point to point we thus go to acquiring several millions of
points per second. With a rate of 60 images per second, a whole disk can be done in a time-span
close to the real time duration of the disk (18O seconds). Thanks to this method, it is not
necessary anymore to have the high resolutions needed in the methods of Carl Haber or Sylvain
Stotzer. In this method, we have enough with 5µm per pixel, considering that it's not the position
of the groove that is being measured. To this, another considerable advantage can even be added: the
height of the side of the groove is sufficiently important to cover multiple (ca. 30) pixels. As the
groove thus seems to be composed by different parallel tracks, redundancy is entering the signal.
Error treatment by the image
Errors that have been established in the grooves may be due to a bit of dust, or a contamination
coming from the exudation of a plasticizer, as is often the case with acetate disks. Errors like
these diffuse the light, instead of reflecting it. In some cases the sides of the grooves are marked
with almost parallel scratched as a result of an abrasion caused by a wrongly adjusted needle or an
excessive pressure. In the case of a directly engraved disk, this phenomenon may be caused by the
harmful effects of a damaged needle. In similar circumstances a calculation of the average depth of
the trace in the total length of the groove may free the result from traces of abrasion. In general
these errors can easily be detected. A cartography of these can even be drawn up to cancel them or
to reduce their effects during the processing of the image, even before extracting the audio signal.
This cartography of errors will eventually help to draw a file with metadata, that during the
digital processing of the signal will provide the exact position of the errors. For these errors
some restoration technique will have to be applied. But bearing in mind that a correction is never
completely neutral, we avoid to tackle the signal where it is healthy. On the longer term, the
system should also be able to carry out an automatic detection and correction of cracks as seen on
the 'marble-lined' disks, as reconstructing these puzzles is no more than a software issue (fig.
14). In the case of a 78 rpm disk, this system offers 240.000 samples per second on the outside. At
the centre of the disk there are still 78.000 samples per second. This means that compared to a
digital signal of 48 kHz, we still have a comfortable over-sampling. The response in matters of
frequency is very good and allows to detect details of a signal down to 15 kHz on a 78 rpm. Of
course, in case of a 33 kHz micro-groove, this sampling is seriously being reduced because of the
slower rotation speed of the disk. This effect can easily be compensated by an appropriate
magnification of the optics of the camera. But, as everybody knows, before ending up with a complete
reconstruction of the audio signal, it is a collection of images of whom a certain number is
deliberately operated as an overlap zone. To cover a 30 cm disk, circa 24.000 images are made.
During this acquisition the camera is positioned on a radius that remains the same during the
complete rotation of the disk. In this way a series of images is made forming a crown. From this
crown the image processing software calculates the junction points of the six observed tracks,
thanks to the overlap zones of these images. After decoding, we obtain thus the equivalent of a
multi-track piece of audio, covering a duration corresponding to the time needed for one complete
rotation of the disk, or 769 ms. In this way the whole disk will be decomposed into crowns of
multi-track audio with a length of 769 ms.
Don't hesitate to follow the gradient
Today this system named Clareety has already been tested at INA and the results are promising. It's
development is part of the PrestoSpace project. In 2006 INA associated with an industrial partner,
INDEEP from Annecy (F), aiming to industrialize and commercialize it. The system responds completely
to the goal it was designed for: no-contact reading-out acetate disks that were recorded from the
thirties to half the fifties, especially those for which the fragile state doesn't allow any
read-out using a needle (fig. 15). The test particularly proved that direct digitization from the
gradient of the grove offered a bigger sensibility of the system than the sensibility coming from
calculating the width of the groove followed by a calculation of its derivative function. But way
beyond this first goal, this system has proven its capacity to read-out 78 rpm's with lateral
engraving that have been commercialized in the fifties, just as the first Berliner disks
characterized by a very particular groove profile. A deviation at 90° of the optical read head
allows moreover to read Pathé-Saphir disks with a vertical engraving. Also the read-out of 33 rpm's
is proven. Inevitably we come to the question of stereo disks. Two solutions are possible. The first
is reading for example the outside of the groove (right channel), then the inside (left channel) and
then synchronizing the two tracks to get a stereo signal. This solution doubles the time to take the
images. The second solution consists out of installing a second optical head to scan simultaneously
the other side of the groove. In that case the system is a little more complex and costly, but also
realizable. To read bent disks (victims of a 'curtain effect') without a hitch, the system will be
equipped with an automatic correction of the altitude, keeping a perfect focalization during the
whole rotation of the disk. Normally, at the end of this year, a first industrial system will be
A new vision on the analog world
Finally we can say that the solution proposed with this INA-system is as far as we know the only
system for optical analysis and digitization of disks that's being industrialized. It's major asset
is the availability on the market of a product that meets the technical and economic requirements of
archiving institutions, as well as those of audio restoration professionals. As far as we
understood, this system would cost little more than an ELP turntable. Of course we're still far from
a system that's affordable for aficionados of rare recordings. But what's important, is that from
now on we can be sure to find a system capable of extracting thousands of hours of recordings out of
these phonographic pieces. There's good hope that finally a lot of archiving institutions will be
able reveal these fragile memories. Thus, for once, the marriage of the digital and the optical
doesn't consign the old analog carriers to the attic, but helps them to assure an eternal life.
*** Disclaimer ***
Vlaamse Radio- en Televisieomroeporganisatie
Auguste Reyerslaan 52, 1043 Brussel
nv van publiek recht
BTW BE 0244.142.664