Restoring Gray Audograph Recordings
30 second sound sample from an Audograph
(Recorded 20 June 1960, Patterson VS Johansen title fight on KWJJ, Portland. Apparently the Audograph microphone was held up to a radio speaker. Yes, that's Howard Cosell at the beginning.)
There are a number of interesting challenges to be overcome when recovering a recording from a dead media. The Gray Audograph is definitely "dead". Introduced in the mid-1950s the Audograph was designed as a portable dictation system using thin plastic recording discs. These discs are about the size of 45rpm records but are as thin as giveaway records found inserted in magazines. Very flexible, they are prone to damage. The spiral groove is also very shallow making it hard for the needle to track correctly on playback.
Lacking an Audograph upon which to play back the original recordings, the restorer has to be fairly ingenuous. The media obviously is "like a record" in that there's a spiral groove in which a needle-like thing travels picking up analog vibrations for amplification. That is almost the only aspect which is obvious.
The first non-obvious problem is the Gray Audograph plays from the inside to the outside of the disc. After putting the needle in the groove a number of times only to hear a single revolution playback before the tone arm jumps off the record onto the platter, it makes sense to start the tone arm a few turns in on the record. Eventually the discovery is made that the Audograph doesn't play from the outside edge to the center of the recording, but inside-out. If the grooves were deeper so the needle didn't skip as often this would be easier to learn. Coping with tone arm weight and tracking are much easier after finding the correct direction of play.
Unfortunately the Audograph record starts far inside the radius of a "modern" lp. This causes automatic turntables to start the end-of-record cycle forcing the tone arm back to the outside to end playback. Even a turntable without automatic rejection won't allow the tone arm to swing so close to the center of the album because this is the space usually occupied by the label. Correcting this frustrating problem is simple with a mallet and pliers.
Now that the needle will run through the entire album without too many problems it's time to learn how fast the platter must spin for correct playback. At 33rpm voices sound close to normal towards the beginning of the recording. Towards the end of each recording, the outside edge, the voices are distinctly chipmunky. The distance travelled by the needle changes from the short-circumference inner track to the longer circumference outer track. Apparently the Gray Audograph, in an attempt to prolong recording time, compensates to maintain a constant needle speed.
To get the desired frequency response at the start of the disc, the Audograph needs to travel at a certain speed, say thirty rpm. But at the outside edge of the disc, this is not needed. The needle will travel the same distance with a significantly lower turntable speed, a difference of over 50%. This is comparable to playing "crack the whip": the inside ice skaters are traveling at a walk, but the skaters on the end of the whip are going very fast. The Audograph attempts to keep a steady needle speed by slowing the recording rate of revolution as the needle goes to the outside. How it does this is not clear. Obvious guesses are it senses the angle of the tone arm or it simply slows a given amount every minute of recording. Unfortunately it's not entirely consistent.
To get an accurate restoration is challenging. Exactly how fast should the playback be at any given point in the recording? There's no benchmark. The human voice sounds wrong at double-speed. It also sounds wrong at half-speed. But to find the exact pitch is difficult. The restoration at hand includes recordings of children, one of whom plays a trumpet. Trumpets are difficult to play. To overcome this, most easy songs a trumpet player first learns are in keys which are simpler to play; usually Bb or F. Assuming the trumpet player is in tune, shifting the recordings to force the pieces into Bb is a good start. In fact,
it's not terribly difficult to tell what note is actually being played by how the beginning horn player reaches that note. Some notes are just easier to play, so they tend to stand out.
Short sections of the digitized recording can be resampled to take advantage of the information learned from correcting the trumpet to its correct pitch. Converting the same radius areas of other Audograph discs demonstrates that voices sound correct also. The guess is probably correct depending on how far out of key the nine-year-old trumpet player was on the recording. As a benchmark, the trumpet is a good start.
But there actually is a very accurate benchmark on each Audograph record. It is unintentional and probably entirely unnoticeable without a very good analytical system. All of the recordings were made in the United States. Electrical power in the US uses alternating current at sixty hertz. The electric motor spinning the Gray Audograph runs off a normal wall plug. A very small amount of the hum of the motor and probably some inductance in the circuits is carried into the
cutting-needle and embedded in the record. Not only does the intended content of the record mask this noise, the pops and clicks are also significantly louder. Seventy decibels louder, in fact. Finding a few milliseconds of "silence" where the needle was cutting the record but nothing is actually happening is tricky. Performing a detailed frequency analysis of the octave around 60Hz there is often a peak one or two decibels above the noise floor. Assuming it's the hum of the motor, it's simple to calculate a ratio to apply to resample the recording so the motor is humming at exactly the right pitch.
The hum is characteristic of all AC circuits. Flourescent lights are the most obvious source of sixty cycle hum. The buzzing sound heard is not just at sixty cycles per second, but at harmonics of the fundamental pitch. Any time a signal is carried through wire it is liable to be affected by AC power hums. Audio engineers spend lots of time trying to avoid it. This may be the only time the hum has ever been useful.
Some harmonics of sixty cycles are also present in the recording consistently, lending credibility to the hypothesis. Applying corrections to various parts of the recording produces pleasing results. For any single disc, the calculations produce seemingly accurate playback. Voices sound good. The trumpet is in the correct key as much as a nine year old is likely to be. By taking an entire seven minute side of an Audograph record and continuously scaling the playback to maintain a steady 60Hz electrical hum, an accurate version is produced. Note that it's clear the hum corrected for is not induced by playback because the rough correction (via trumpet) moves current era hum out of range.
Unfortunately, the correction scaling factor is slightly different on each record. Each record begins and ends at a nonstandard radius. While the needle may have been dropped in "the same place" every time, there is enough variation to frustrate an automatic scaling conversion. This probably affected playback on the Audograph itself and caused minor variations in the pitch of playback. The restored product will actually be significantly better than an actual Audograph device in maintaining correct pitch.
The end of each record is whenever the person recording noticed the record was almost out of space. When adults ran the machine, album sides average about seven minutes. When children run the machine an album side is sometimes almost nine minutes until the cutting needle literally dropped off the record. The outside edge of the record is poor in quality both for pitch (rotation speed) and signal-to-noise ratio. There is not a clean sixty cycle hum from which to learn the correct turntable speed so it's just a guessing game as the noise gets worse and worse towards the end of each record.
The end results of restoring recordings from a Gray Audograph reveal a signal-to-noise ratio (best case) of about 15dB and a frequency reponse at the high end of no more than 5kHz. The low frequency response is difficult to discern. There is plenty of content down to the subsonic range, but it's not useful. Since there's a clear component at 60Hz the Audograph obviously was able to record there even if the microphone attached to the Audograph wasn't capable of capturing a frequency that low. In fact, this may explain why the sixty cycle hum is able to be discerned at all; there is no other component reaching the cutting needle from the microphone at that frequency. The recording is monaural
and there is significant crosstalk when the system was overdriven, but that's not terribly interesting.
With variable pitch corrections made specifically for each album side, all that remains is to equalize out some of the modern era inaccuracies. A roll-off above the frequency response of the original recording seems safe as there's obviously no component in the highest two octaves. Below sixty hertz there's plenty of mechanical noise which just thrashes speakers for no good reason, so the low end is rolled-off also. Bringing the signal up to listenable levels while crunching the pops and clicks is also useful. Finally, a conversion to stereo at 44.1kHz for placement on a compact disc.
It's worth noting in another fifty years the compact disc will have deteriorated into unplayability while the original vinyl Audograph recordings won't have appreciably decayed. Even if the CD were still in good condition, trying to discover how it was intended to be played is a much more difficult task than learning how the Audograph worked.
La Honda, CA
2 July 1999
[I've also worked with "Recordio Discs", "Tru-Kut", Edison "Voicewriter" and a few other odd formats.]