Interview: Carl Haber

Seeing voices

Print edition : April 17, 2015

Carl Haber in his recording research lab with the IRENE machine at the Lawrence Berkeley National Laboratory in Berkeley, California. Photo: Roy Kaltschmidt/Berkeley Lab

An Edison cylinder phonograph, circa 1899, using wax cylinder to record and playback sounds.

A Victor V phonograph, circa 1907. Flat disc phonographs began to displace cylinder phonographs in the 1890s.

Brown wax phonograph cylinders.

Interview with Carl Haber, a senior scientist in the Physics Division of Lawrence Berkeley National Laboratory at the University of California.

SOUND was first recorded and reproduced by Thomas Alva Edison in 1877. Until about 1950, when the use of magnetic tape became common, most recordings were made on mechanical media such as wax, foil, shellac, lacquer and plastic. Some of the older recordings contain material of great historical interest but are found in obsolete formats, and are damaged, decaying or are now considered too delicate to play.

Unlike print and latent image scanning, the playback of mechanical sound carriers has been an inherently invasive process. Recently, a series of techniques based on non-contact optical metrology and image processing has been applied to create and analyse high resolution digital surface profiles of these materials. Numerical methods are used to emulate the stylus motion through such a profile in order to reconstruct the recorded sound.

In this interview, Carl Haber, an experimental physicist, who has put together the IRENE [Image, Reconstruct, Erase Noise, Etcetera] explains this approach. Carl Haber received his PhD in physics from Columbia University and is a Senior Scientist in the Physics Division of Lawrence Berkeley National Laboratory at the University of California. His career has focussed on the development of instrumentation and methods for detecting and measuring particles created at high-energy colliders, including at Fermilab in the United States and at CERN (the European Organisation for Nuclear Research) near Geneva, Switzerland. Since 2002, he and his colleagues have also been involved in aspects of preservation science, applying methods of precision optical metrology and data analysis to early recorded sound restoration. He is a 2013 MacArthur Fellow and a Fellow of the American Physical Society and the John Simon Guggenheim Memorial Foundation.

IRENE is an application that demonstrates the coming together of the humanities and the social sciences. How did you get this idea of preservation of sound and what prompted you to develop this equipment?

My career has been in studying fundamental particles and with electronic experiments at CERN, and my particular interest lies in how we measure particles and how we understand their properties from an experimental point of view. So, I have always been thinking about images and signals and how we process them in an automatic way and how we get information out of pictures. Very often in physics, we get interaction of matter and particles as a way of getting our information. I have always been thinking about pictures and also how to use computers to get information from pictures. So, one day, I heard about the problem of sound recording collections and how materials in these collections could be retrieved. [These materials are] sometimes damaged, sometimes very delicate.

One of the big problems is that when you try to play them, you need to put a needle in contact with the surface and that is invasive and will damage it further. It will be impossible to play if the material is broken. Since I was thinking a lot about pictures, I got the notion that maybe we can get really good pictures of sound recording materials or records and then we could get information from those pictures rather than by putting a needle in contact with the material. We found some very powerful tools and we found a very good way to combine these together and get pictures of sound; in fact, we got some pretty good information from material which otherwise is not playable. I was very happy to find a way to use science to benefit the humanities.

From where did you hear about the sound preservation problem with the records?

I work with the Lawrence Berkeley National Laboratory which is in Berkeley, California, near the campus of the University of California. It’s about 50 miles from Silicon Valley, which is a big centre of microelectronics and information technology. In my work, I spend a lot of time going back and forth from Silicon Valley and work with many small companies with different types of electronics that we need for our research.

When I was driving back one day from Silicon Valley, I was listening to the radio, and there was a report about the sound collection at the Library of Congress and about how there were all these problems with materials that were old and delicate. The report was an interview with the famous American musician Micky Hart. He was a member of a famous rock band [The Grateful Dead] and very involved in the world of music as well. One of his interests is to help preserve sound recordings. He was arguing that this was an important issue, writing books and articles about it, and so forth. I knew who he was from his music career.

After I got back, I wrote things down in my notebook and I started to look around on the Internet and did some research on how sound recordings really work. I also found some books to educate myself in this subject.

Can you describe the IRENE system? How does it work and what is the process involved?

IRENE is designed to restore sound from sound recordings that have a groove. For example, a magnetic tape doesn’t have a groove, a CD doesn’t have a groove; they are optical devices and we don’t work on those. IRENE is only for materials that have a groove. It works on devices like phonographic records, which are familiar to people, wax cylinders, which are like phonograph records but with a different configuration, and then some strange recordings on a variety of materials which were experiments in the very early years. So, we have cameras, special scientific cameras with special optics and illuminators which let us get very sharp and magnified images of the surface of the sound recordings. We can then run these images through the computer to extract the information from them about where the grooves are at every point around the record or the cylinder. Once you know where the grooves are and how they move, it is simple to make out what the sound is.

Of course, this is based on pictures; we can also manipulate or change the picture in any way we want. Now, we are not going to change it in some silly way, but if there is a scratch or a broken plate or dirt or some other defect, we can actually take it out of the picture if we want. Then, we fill in digitally what missing. We smooth it over and make a continuous motion. That’s a little like using Photoshop to remove a scratch, a particle, a blemish or a red eye in pictures. You take the part you don’t want out and you use the parts around it to create an appearance of improvement. We are not really rescuing lost information; we are just making the picture better or making it easier to listen to the sound.

If you are a linguistic researcher, for example, trying to understand what is being said, sometimes taking away the noise, which can be due to scratches or dirt, makes the sound easier to understand. So, first we take pictures and get information from the pictures; the technical name for this is imaging.

You said you restore sound from the records. What kind of audio standards are being used? Are very high resolutions needed?

Sound exists in time and also exists in frequency, or pitch. A person with good hearing can hear anything up to 20,000 cycles per second. In western music, the standard pitch, the A above middle C, is usually set at 440 hertz; the frequency of the electric mains in the United States is 60 Hz, it is 50 Hz in India. These are all different frequencies. So, in order to have a sound recording that is properly digitised, you want to make sure that you have all the frequencies that are there. The specification used at the audio unit is that the sound recording could have a sampling rate A frequency of 96,000 Hz. It is almost four times the highest frequencies that we can hear. The CDs that you buy in the shop have music encoded at 44,100 Hz, which means you can get any sound up to 20,050 Hz. It really is the limit of human hearing anywhere, but IRENE can easily measure more than a 100,000 samples per second, so we can match specifications that the audio community asks for.

Who is using the system now?

There are five IRENE machines in the world today. The one at Berkeley is our research and development machine. There are two machines at the Library of Congress in Washington, D.C.; then there is a machine at the Northeast Document Conservation Centre, which is near Boston, Massachusetts. This is where people can go with their recordings, say, from a library or museum, that need to be digitised. They do it as a service. The fifth machine is here in India at the Roja Muthiah Research Library in Chennai. Now, the IRENE machines can do two-dimensional imaging or three-dimensional imaging. Some of the machines can do both, but some machines can only do two-dimensional imaging.

What do you mean by “two dimensional” and “three dimensional”?

The phonograph record is a very common format. There are probably 500,000 different recordings made on phonograph records, in India and South Asia. In the world, there are probably over a million or more than a million recordings on phonographs. The grooves in the phonograph move from side to side. When we talk about two-dimensional imaging, we mean using a scientific version of photography, something which can measure a groove from side to side. We only need to measure the two dimensions (side to side) in that application, so we call it two-dimensional imaging.

In a [wax] cylinder or some other material, the sound is actually moving up and down instead of from side to side. Imagine flying over a mountain and you need to know the height of the mountain and the depth of the valleys. Photographs taken from the sky do not give you that information. You have to use a special kind of imaging that also measures the third dimension of depth and we call that three dimensional. So to do that, we use a special kind of microscope that is part of the IRENE system, where it is configured to read and measure the dimension that is necessarily for cylinders and other particular medium.

Is this system or equipment available for sale?

We built them one by one in the laboratory as scientific devices and IRENE has not been turned into a commercial product.

What are the advantages of this system?

The main advantage of this system is that you can play things that are unplayable. So records that are broken, in really bad condition, material which is too delicate to play and very soft material can be played with IRENE. Material for which you don’t have a playback machine anymore can be played in this way because we can essentially simulate what the playback machine did in the software.

For vertical recordings, we can actually get more sound off the surface than you get by playing them with the needle. So IRENE has a very clear advantage for things like wax cylinders which have vertical grooves.

Another advantage is that you can easily turn this into a mass production system. You can set it up in such a way that a lot of discs can be processed in a work flow that will avoid some of the problems that you get with scratched, broken or damaged discs. You can just scan them; the process does not stop for the problematic item.

Can you play all kinds of records, or is there any restriction on what can be used with this machine?

In principle, you can play anything though you will not always get a better playback in this method than with the stylus. For example, if you have a stereophonic LP from, say, 1960, playing it with a needle in the normal method will generally be better. But, as the material starts to degrade or gets broken or damaged, it is more advantageous to go with IRENE.

Lacquer disc, for example, is a type of disc which was used in recordings in radio transcriptions and interviews from the 1930s onwards. They have a thick coating of lacquer on glass or aluminium. Sometimes the lacquer can crack and it is impossible to play it with the needle. It is really hard to play such material with IRENE too when many pieces start to move around, and it gets hard to keep them in focus. Over the next year, one of the things we will be working on is some new techniques and tools to better master the lacquer disc. We played some really good lacquer discs but we want them all to work without the intervention of an operator.

Is there not another hurdle in the system? While you are restoring these records, you also end up filling up huge volumes of images. What do you suggests for archivists?

I think the images are large, but the cost of digital storage has been getting cheaper. I think you should archive the images on large disc drives. They don’t cost huge amounts of money. My suggestion would be to archive the images according to an accepted standard.

What are your future plans for the IRENE project?

Our plan is to try and convince people that we should do a very large project. One such project that we are looking at very seriously is the decoding of recordings contained in over 3,000 wax cylinders. These contain recordings of native Americans, mostly from California, that document the language, culture, life and stories of their society. These were recorded between 1901 and 1916. We have already submitted a proposal to the University of California, Berkeley. We want to use the 3D Irene scanning system and digitise the entire collection.

Another interesting collection, at the Smithsonian Institute in Washington, has over 200 unique and unusual recordings that were made by Alexander Graham Bell and other people in the 1870s. We probably need a few years of work to digitise everything in that vast collection in a professional way and I would really like to mount a dedicated project to digitise the whole collection, which I’d call a history of science collection. It documents a process of invention in the late 1870s.

Sundar Ganesan is Director, Roja Muthiah Research Library, Chennai. He has a postgraduate degree in history from the University of Madras and a degree in library and information science.

This article is closed for comments.
Please Email the Editor