5. 'AUGUST 2026'

My position at the Workshop had by this time become well established and, as was traditional in large organisations, I had been 'volunteered' to go on a two-week management course*. My colleagues joked 'He's being measured for a suit!'
[*At the time, all those on certain grades, even those without staff, had to go on management courses.]

Unfortunately, I only had two days to play with the new 'synthesiser' [the EMS Vocoder 5000], but the results were encouraging. What impressed me most was that if I took a clean recording of a voice, the re-synthesised sound could, with care, be very close to the original. In fact, the re-voicing could be clearly identified as belonging to a specific person. Unfortunately, I didn't have time to find out what would happen if I altered the control voltages. I had to attend the course in the Langham. Although the course was up to the BBC's high standard, the vocoder was constantly in the back of my mind.

Attending the course though, turned out to be useful. I was accustomed to driving between home and the Radiophonic Workshop in Maida Vale. As there was no parking available at the Langham, I had to travel by train. The journey took twenty minutes, and I decided to buy a paperback to help while away those few minutes. I chose a collection of short science fiction stories by Ray Bradbury called 'The Martian Chronicles'. When I read the story 'There Will Come Soft Rains', I knew I'd found something very special. It was an ideal vehicle to test out the vocoder.

Within a few days, I had completed the script for a radio drama which I called:- 'August 2026: There Will Come Soft Rains, for Narrator, Vocoder and Synthesiser.'

Our jobs at the Workshop had just been re-graded and we were now called 'Producer/Composer Radiophonic Workshop'. This was useful as it gave me access to the 'Drama Rep'* pool, a group of actors who were available for half a day, to work for all production departments, without cost.
[*The BBC Radio Drama Company (RDC)]

RECORDING

I decided that I could do the whole drama at the Workshop. Garard Green was booked as the narrator for the Monday morning of my return to the Workshop and Eva Haddon and Jeffrey Segal were booked for robotic voices on Tuesday morning.

Recording the narration first with Garard Green was a good idea. Garard is one of the best narrators in the business and what I wanted from him was a detached and uninvolved commentary. He asked if there were was any of the programme recorded that he could hear, so that he could get into the mood of the piece. I explained that there wasn't, and that I thought this would in any case be unhelpful, as what I wanted was the feeling of a narrator standing outside the scene, looking in, but uninvolved. The audience's involvement was to come from their interpretation of the sounds of the house.

Garard understood this perfectly and recorded all the narration in one take.

Recording the voices was more difficult. Both Jeffrey and Eva were determined to be robots. I explained to them that by the year 2026 synthesised voices would be very sophisticated, and what was wanted for the drama was real character and human qualities for the barometer, calendar, clock, etc. They took this point fairly well, but it wasn't until I demonstrated their recordings through the vocoder that they really got into it.

Using the vocoder was not easy, as all the controls needed very careful adjustment. In most cases only one setting across a range would work, and slightly moving just one control would spell disaster. Our patience was tried by the vocoder, but it was important that we went through this demonstration process, not only to clarify the nature of the actors performances, but for technical reasons as well. The voice recordings had to be made very carefully. Any extraneous sounds, like script rustles or unwanted mouth noises, would be translated by the vocoder into speech-like sounds, which were very confusing. One thing I hadn't realised at the time was that any acoustic would also add confusion, and that a close microphone technique with some bass-cut was essential. I used an AKG C60 cardioid microphone with windshield at a distance of 10cm.

Plosives [consonants that are produced by stopping the airflow using the lips, teeth, or palate, followed by a sudden release of air] and unwanted mouth sounds, unavoidable with close microphone recording, were later physically edited out of the tape.

The next day was spent cleaning up the recorded voices by editing out, with razor blade and jointing tape, all unwanted sounds. It was amazing how many of these there were.

VOCODING

Then came the exciting part - vocoding the voices. I decided to make the 'house calendar' voice first, as what I wanted was a treatment as close as possible to the original voice, but with just a hint that it was produced electronically.

Eva performed this:

The recorded sound was fed into the vocoder's input (see above block diagram) and its level adjusted to peak to '6' on the peak programme meter (PPM) [equal to +8dBm]. This level was critical. If it was too low the analysis systems failed to work accurately and if it were too high distortion would quickly set in. The sound was then routed via an analyser switch.

Analysis

As speech consists of pitched elements which have a fundamental frequency and noises, without a recognisable pitch, the analyser switch looked at the input, and decided, at any one moment, was it detecting a sound or a noise? It did this by filtering the sound progressively down through a low pass filter until it found a fundamental frequency.

• If one was found, a sound had been detected; the oscillators were switched on and the white noise generator turned off.
• If no fundamental was found, the input must be a noise and the oscillators were turned off and the noise generator switched on.

A balance control allowed the ratio of sound to noise switching to be adjusted, and this could be set to an extreme where only one or the other was permanently on.

When noise only was selected the result was a sort of whispering. The sound 'permanently on' mode was most useful when feeding in a complex external sound, to make the voice of a talking concrete mixer, for example.

Spectrum and Synthesis

The input sound was also routed to a spectrum analyser. This generated nineteen control voltages which were derived from band pass filters set at:

There is a high pass and low pass filter at the band range ends. As the human voice generates its pitched sounds with a pair of vocal chords, the waveform produced is not unlike that of a double-reed musical instrument [such as an oboe or bassoon].

This waveform is called a ramp because of its shape, as shown below:

It can be viewed as a consisting of two components; a sloping section which emulates part of a triangular waveform and a vertical section, part of a square wave. This combination results in a very rich harmonic structure, which has both odd and even harmonics, and is ideal for the 'subtractive' process of synthesis the vocoder uses.

The Oscillators

Two ramp oscillators are used as the pitch generators of the vocoder, and they are voltage-controllable. This is important, because speech is modulated. We do not speak in a monotone. Much of the expression in what we say comes from the fact that our voice is articulated (goes up and down in pitch).

A pitch to voltage converter after the sound/noise switch detects the change in the fundamental frequency of the incoming sound, and this controllable voltage is fed to the control input of the oscillators. The voltage can be turned off, and an external CV can be used instead, which can be derived from a musical keyboard. In this case the 'vocoded' voice can be made to sing.

The Output

Finally, the combined, switched outputs from the oscillators, white noise generator and external sound source are fed into the voltage controlled filter. This has twenty-two controls, [some of] which can be seen in the photograph [at the top of this chapter]. Nineteen of these correspond to the spectrum analyser frequencies, with the addition of a high pass filter above 8kHz, a low pass filter below 200Hz, and an additional control of the overall volume. The analyser outputs are connected to the filter via a twenty-two-way matrix. Any voltage out can be connected to any control input by a linking pin placed in the matrix.

VOICES

At the beginning of the story Bradbury describes a landscape after a nuclear holocaust. I decided to use the vocoder's raw sounds to depict this chaos. Gradually out of the chaos a single voice of intelligence emerges - the robot voice of the calendar speaking the date in the only house left standing. As the raw vocoder sounds assemble the landscape, first we hear the rhythmic switching choice between noise or pitched monotone ramp wave.

Then as pitch change is detected the voice gradually becomes fully articulated:

The landscape voice is further refined. The pitched information is placed alternately left and right and, as the vocoder's parametric filter becomes more active, the voices move together and combine. This is the landscape sequence, as used in the opening and again at the end of 'August 2026', in reverse, to depict the loss of all traces of Man's influence.

The following example is the untreated voice of Jeffrey Segal, as the house clock and is for reference:

The vocoded voice of Jeffrey Segal as the house clock: the pitch to voltage coverter output is changed to raise the pitch of his voice and the noise content is increased and sharpened in the filter output mix to give it a metalic feel.

Clock mechanism, vocoded voice: I discovered when making this clock voice that when I spooled back the tape by hand, the vocoded output sounded very much like a clock mechanism. This time-lapse section was achieved by reverse hand spooling the master tape.

For this clock montage two clock voices were used with their pitch range gradually increasing. Tape delay, timed feedback, was added to emphasise the rhythmical content.

Here the voice of Jeffrey Segal is vocoded with the ramp oscillator set to a lower frequency than detected by the pitch to voltage converter, and the low and medium high frequencies boosted by 6dB. This helps to give him the shape of a large 'banjo' type barometer.

The house is on fire: The calendar, feeling the effects of heat, tries to perform a Tennyson poem. Here no pitch information is sent to the excitement oscillator's control voltage input, with the result that the voice speaks in a monotone. The output filter's envelopes are also set with long decays, so that the words become slurred and run into one another.

The clock also feels the effects of the heat.

The computer, also feeling the heat, is reminiscent of 'HAL' [the computer] in '2001'*. The pitch of the exciting oscillator is set unnaturally high and an external control voltage is fed in from a keyboard to make what was originally only spoken become the famous music hall song ['Daisy, Daisy']. At the end, the pitch change is levelled out.
[*'2001: A Space Odyssey', a 1968 epic science fiction film produced and directed by Stanley Kubrick.]

Jeffrey Segal recorded several fire alarms. In the process of vocoding, the high frequencies were preferred and slightly distorted to help give a feeling of the searing heat. Also some of the detection controls were set near their threshold, so that they worked spasmodically, enhancing the feeling that this situation might be more than we can cope with.

A collection of Eva Haddon's fire alarms:

SOUNDS

As well as the voices, many other sounds, heard in and around the house, were required.

Here are some small examples.

The cooker: This is made basically from tuned white noise set around a frequency range to emulate the feeling of escaping steam.

Eight pieces of perfectly browned toast: Here the pitch control voltages were fed through a slew limiter. This controlled the rate of change of the incoming voltage over a period, which could be set to be instantaneous or take several minutes. Here the limiter is set at about one third of a second. This gives the impression of the toast springing up out of the toaster.

Rain Drops: For these the slew limiter was used again. Pitch control voltages from a random voltage generator were fed through a slew limiter set at about one tenth of a second, but this period itself was voltage controlled by an input from another random voltage generator.

Bridge Tables: This is a music section, made on the Delaware synthesiser, to cover an interlude where card games are being played. The middle section of this music is a retrograde inversion of the main theme, made by driving the sequencer 256 in reverse.

TINY ROBOT CLEANING MICE

For this section, the first thing I did was write a little waltz (shown above) to accompany the mice on their cleaning escapade.

Each note needed to be shaped and have the 'cleaning mouse' characteristics described by Bradbury. This was done on the Delaware synthesiser. We are told that these mice are electrical, and so I made the assumption that they would operate and sound like miniature vacuum cleaners. The Delaware's envelope shapers were rudimentary by comparison with a modern synthesiser, although there were six of them and they could be combined to produce some more interesting shapes. For this requirement, however, the basic shaper was ideal.

There were three controls: Attack, Sustain and Release (ASR) The Attack controlled how quickly the sound output reached its sustain level. Sustain controlled how long the note was held, and Release set how quickly the note would fade to silence (no output).

If the Attack and Release controls were set to zero and Sustain set to half way (control value of five), then the shaper worked like a switch, as shown here:

EMS called these envelope shapers 'trapezoid generators', because when the Attack and Release controls are brought into play the shape of the output takes on a form like the geometric shape the trapezium (a quadrilateral with two faces parallel).

I decided that the basic mouse sound should be filtered white noise, set at a fairly high frequency, appropriate to the scale of the mice. Because the mouse sound was probably being produced by an electric motor, the shape of the sound should follow the rules of electrodynamics. The motor would reach its operating (constant) speed quite quickly, due to the electromotive force. I set the Attack to about half a second. The sound was sustained for one second and the Decay set to four seconds: the decay being longer than the attack, due to the inertia of the motor and absence of any electrical involvement.

[The dope sheet, showing the settings for the VCS3, is shown below:]

When we hear a motor run up to speed we expect it's pitch go up and likewise when it stops and runs down its pitch to descend. Fortunately, the shape of this pitch rise and fall follows the same shape as the sound level envelope, which is in this case defined by the trapezoid controls. The envelope shaper actually generates a control voltage, which follows the shape defined by the ASR controls. This voltage is applied to a voltage controllable amplifier and this actually shapes the sound. If we simultaneously apply this control voltage to the filter frequency control input, the filtered white noise will rise and fall in pitch in synchronisation with the envelope shape.

Finally, to give the impression of the mice 'whirling their moustached runners', some pulsed square wave was applied to the filter frequency control. As this could again relate to an electrical idea, I chose a frequency that we might expect the carbon brushes to make as they contact the comutator [at the centre of a motor]. This is normally fifty cycles per second (50Hz), but as the square wave is made up of two pulses in each cycle, what we hear is a one hundred cycle ripple.

[The following video shows how Malcolm created the 'cleaning mice':]

THE PROGRAMME

Editor's Notes:

The full script can be viewed by clicking here. Click the 'Back' on your browser to return.

The programme won the Society of Authors Award for Radio Drama and the Imperial Tobacco Award , 1977 (now the Sony Award).

Below is a letter from Ray Bradbury, praising Malcolm's work. Unfortunately, no-one took up the opportunities he suggested.

©Malcolm Clarke 2003