ISA Interchange

Echoes of the Past: Restoring a 1946 Voice Through Signal and Image Processing

Written by Steve Mustard | Mar 14, 2025 11:00:00 AM

I’ve always been interested in image and signal processing. It was a big part of my final year project at Sunderland University (using the Kalman filter for condition monitoring in the presence of noisy measurements) and for a lot of my career at Logica (processing infrared imagery from reconnaissance aircraft, detecting enemy submarines from passive sonar signals and speech recognition in the noisy environment of the Eurofighter cockpit — not that I can claim to have been the brains behind any of those innovations!).

Signal processing is a fundamental component of closed-loop control systems, ensuring accurate and reliable operation across various industries. Sensors inherently pick up noise from environmental factors, electrical interference and mechanical vibrations, which can distort measurements. Signal processing techniques such as filtering, averaging and spectral analysis help remove unwanted disturbances, enhancing the precision of feedback to controllers and reducing uncertainty.

For example, in industrial automation, temperature sensors in a manufacturing plant may experience fluctuations due to transient heat sources, but signal processing techniques like Kalman filtering can extract the true temperature trend for better process control. In automotive applications, anti-lock braking systems (ABS) rely on real-time signal processing techniques such as Discrete Fourier Transform (DFT) to filter out road surface irregularities, ensuring smooth and responsive braking. In aerospace, gyroscopes and accelerometers in aircraft navigation systems use signal processing such as Discrete Wavelet Transform (DWT) to eliminate noise caused by vibrations, improving flight stability and accuracy. Even in biomedical engineering, heart rate monitors and electrocardiogram (ECG) machines apply signal processing algorithms such as Hilbert Transform to remove muscle artifacts and electrical interference, providing clearer diagnostic readings.

I'm always fascinated by advancements in signal processing from diverse fields and their potential applications in control systems. A recent personal project provided me with fresh insights that I found particularly intriguing.

A few years ago, when my mother passed away, I collected some old photographs and memorabilia. Among them was a Voice-O-Graph disc. The label said it was recorded by my father, Johnny, in New Orleans in October 1946. He had passed away several years before my mother, and so much of the history of this disc was lost.

The Voice-O-Graph disc from 1946

The Voice-O-Graph is an innovation that you would never have realized had existed when it did. It was invented by a European engineer named Alexander Lissiansky and patented by Mutoscope in the US in 1940. A version of Voice-O-Graph was exhibited at the New York World’s Fair in 1939. There are very few Voice-O-Graphs in existence today, but there is a Facebook group and a website where you can learn more.

1940s-era advertisement for Voice-O-Graph

Looking at the label of the disc, my father must have used a Voice-O-Graph like the one shown below to record a message, like an audio postcard. He was in New Orleans at the time, presumably during one of his voyages as an engineer on a merchant ship.

1946 era Voice-O-Graph

There’s a video that shows a working Voice-O-Graph from the 1946-7 era. As automation professionals, you can marvel at the automation involved in cutting the disc with the audio recording while you wait!

The 1940s was a significant era for automation, and the Instrument Society of America was formed in 1945 to support standardization and knowledge sharing. The Instrument Society of America would later become the International Society of Automation, recognizing the global significance of this profession.

The Voice-O-Graph disc is six inches in diameter and has a cardboard substrate with a lacquer surface. These discs were never intended for long term use, and after more than 70 years, mine is unplayable using a conventional turntable. There is some delamination of the lacquer and crazing across the surface. I kept the disc as a memento, but didn’t think more of it.

Several years later, in a reorganization of my office, I came across the disc again. I thought of one of my early mentors at Logica, Dr. Donald McLean. I recalled his groundbreaking work through the 1980s and 1990s where he used computer-based techniques to restore video from discs created more than 50 years earlier.

You may have read that and thought it was a typo. How could there be such a thing as a video disc from the 1920s? Scotsman John Logie Baird made the first public demonstration of a mechanical television system on 26 January 1926. Video disc technology did not appear commercially until the late 1980s (e.g., LaserDisc).

In fact, John Logie Baird made some early video recordings on disc in 1927 and 1928, which he called Phonovision. Some other attempts were made in the 1930s to record BBC television signals to aluminum gramophone discs. Dr. McLean recovered the video from these discs. You can read more about this incredible feat in an article and also in his book, Restoring Baird’s Image.

By pure serendipity, Dr. McLean emailed me around the time I came across the disc again. I told him about my disc, and we talked about methods to recover the audio from it. Several services exist to recover audio from damaged vinyl, and we considered those. Then we came across Project IRENE.

Project IRENE was developed by Dr. Carl Haber at Lawrence Berkeley National Laboratory in Berkeley, California. Dr. Haber got the idea after hearing a report on how historic audio recordings can be so fragile that they risk being damaged if someone plays them by dragging a needle over their surfaces. Dr. Haber used methods derived from work on instrumentation for particle physics, acquiring images and then using image processing to recover audio and reduce noise. In particle physics, detectors are used to track and analyze the paths of subatomic particles, such as electrons, protons and neutrinos. These detectors generate large amounts of data that require precise imaging and signal processing techniques.

The name IRENE comes from the fact that the first audio they recovered was a 1950 recording of “Goodnight, Irene” performed by the Weavers. It is also an acronym for “Image, Reconstruct, Erase Noise, Etc.”

I contacted Dr. Haber, and he connected me with the North East Document Conservation Center (NEDCC), a nonprofit organization based in Andover, Massachusetts, specializing in paper-based conservation, digital imaging, audio reformatting and preservation services. They have IRENE technology on site and provide recovery services to individuals.

I sent them the disc and after a couple of weeks I received it back, along with a set of files.

NEDCC used the IRENE process on my disc. They created a three-dimensional image of the surface of the disc using a technique called confocal microscopy. This captures the depth and horizontal movement of the grooves in the disc in the image.

One of the challenges was not so much the damage to the surface of the disc, but the low level of modulation of the voice signal. You can see this in the image below, which shows the grooves of the disc with very little wiggle compared to a similar disc with better voice modulation. Note how the black lines in the lower image show a more distinct zig-zag pattern.

Image of my disc, showing limited wiggle in the grooves

A better example, showing distinct wiggle in the grooves

Next, the image was analyzed using methods that emulate the motion of a stylus. By calculating the motion of this virtual stylus through the image map, they reproduced the audio content.

Analyzing the image to create audio content

Next, the audio output was equalized. This adjusts the balance of different frequency components in the output, boosting or attenuating specific frequency ranges to improve the resulting quality. In this process, historically informed settings are used to match what would have been typical at the time of recording.

Finally, additional processing was applied to remove clicks and pops and attenuate extraneous crackle or rumble, and an mp3 file is produced from that stage.

When I downloaded the MP3 and pressed play for the first time, the familiar crackle of a needle on vinyl filled the air, transporting me back in time. Beneath the static, the faint strains of New Orleans jazz played softly in the background, setting the scene. Then, through the music and the years, I heard my father’s voice — just as it was over 78 years ago. It was a moment of connection across generations, his words echoing from a past I had never experienced but could now feel.

With the low modulation of the voice and the background noise, it was difficult to make out every word of the 60 second recording, but the gist of it was as follows:

"I'm on the main street in New Orleans, that is, Canal Street... I've been here since Tuesday...Thursday now, we leave tomorrow…I expect…and when we leave there, I'm going to New Zealand. From there I don’t know whether we’ll... but I should think we will, and if we do, I'll be back home early in the new year. I’ve had a good time so far and it’s been…this year, I’ve enjoyed every minute of it. I hope you don't worry about me, because after all, I can see that she’d be worried about you. I hope everything's OK at home and the folks are alright...Ellie, Cissie. Ellie, I missed you. I hope all those boys are ok, and I hope they’ve all got jobs by now. I expect I’ll be seeing them... and you can tell them that...might be more than…but don't worry about me, please."

I don’t have a photo of my father in New Orleans at the time, but his mention of New Zealand is confirmed by this photograph, showing him and a colleague in Queen Street, Auckland on December 13, 1946, a few weeks after recording the Voice-O-Graph disc.

My father (left) with a colleague in Auckland, weeks after recording the Voice-O-Graph

This journey of recovering my father’s voice from a fragile, decades-old recording was a testament to the power of signal and image processing in preserving history. What once seemed lost — trapped in the grooves of a deteriorating disc — was brought back to life through the same principles that have driven my career in engineering. The fusion of historical ingenuity with modern technology allowed me to hear a voice from the past, connecting me to a moment in time I never thought I’d experience. It’s a reminder that innovation not only shapes the future but also has the power to reclaim the echoes of the past.