EVP Recorder

Electronic voice phenomena is what you get when you record a quiet room and then listen to the recording far more carefully than anyone listens to anything.

A quiet room is not quiet. There is the microphone hissing, the building settling, traffic two streets away, your own breathing, the fabric of your sleeve. Slow that down, raise the gain, and there is an enormous amount of structure in it. Some of that structure will, inevitably, land close enough to a word.

Instrument 09 records to your device, then lets you crawl the result a region at a time with the gain up. It marks nothing for you as speech and interprets nothing for you, which makes it slower and less exciting than the software sold for this purpose. That is deliberate. An instrument that tells you what it heard has stopped being an instrument.

Most of what you find will be you. The Survey considers this the interesting part rather than the disappointing one.

The chain makes the noise floor audible. That is the whole operation, and it is exactly what every piece of EVP software does, whatever else it says it is doing.

The lift

A compressor with a very low threshold and a high ratio, followed by a great deal of make-up gain. Everything above the threshold is held down; the make-up gain brings the whole recording back up; and what rises is the material that was below the threshold to begin with. A room that was thirty decibels down is now in the middle of the picture, along with everything in it.

This is not a mix decision on this instrument. It is the instrument, and the control is labelled for what it does rather than for what it is.

The band

Speech is 300 to 3400 hertz — the telephone band, which is where the phone companies found the whole of speech intelligibility to live. Throwing away everything outside it removes spectrum a voice could not have been in.

Narrow is tighter than a voice needs, and it produces the strongest effect. It is included because it produces the strongest effect, and labelled as the least honest setting for the same reason. Low sits under the voice, where building noise and traffic and handling live. Flat is what the microphone recorded, gain-staged only.

Reversal, and the transport

Reversal is included because the practice has a long history with it and because reversed speech genuinely does produce the effect strongly — a reversed recording has all the acoustic texture of speech and none of its structure, which is close to ideal conditions for the ear to invent some.

The transport control resamples. Pitch moves with speed, exactly as it does when you slow a tape down, because that is the honest behaviour and it is what operators of real machines heard.

What a mark is, exactly

A region is marked when three things are true of it at once:

• It is above the local floor. Not above a fixed level — the floor is tracked across the recording, because a kitchen and a stone church have floors twenty decibels apart and a fixed threshold marks everything in one and nothing in the other.
• For the length of a syllable. Ninety milliseconds to about two thirds of a second. Shorter is a click: a knuckle, a phone case, a swallow. Longer is a passing car.
• And the level moved while it did. This is the condition that earns its keep. Without it, the most heavily marked thing in every recording ever made on this instrument would be the fridge — a steady hum sits above the floor indefinitely and satisfies the first two perfectly. Speech modulates; that is most of what distinguishes it from a tone.

The floor is drawn on the waveform as a dashed line. Once you can see it, the rule stops needing to be explained.

A mark is not a detection of speech, and the panel prints that sentence next to the count every time the count appears. Your ear is far better at this than the rule is — which is, in a sentence, the whole problem with the field.

Put the instrument down. A phone held in a hand records the hand. Every adjustment of your grip is thirty decibels above everything else in the room and it is the single commonest thing operators spend an evening reviewing. Set it on something nobody is going to lean on.

Watch the meter before you commit. Arming opens the microphone and starts the meter without recording anything. If there is something running in the room you will see it, and it will be the loudest thing in the review whether you noticed it now or afterwards.

Ask, then be quiet for the whole take. Thirty seconds is longer than it sounds and the urge to fill it is strong. Resist it: a take with commentary in it is a take you cannot review, because every mark will be you.

Note what happened, out loud, at the start. A radiator that clicks, a fridge that cycles, a road outside. Then you know what you are listening to when you find it, which is the difference between an evening’s work and an anecdote.

Review at the lowest lift that lets you hear anything. The temptation is to run it at full and it is a mistake — at full lift everything is marked and nothing means anything. Start low and come up until structure appears.

Listen to the marked regions, then listen to the gaps. The rule is crude and your ear is not. The regions are a place to start, not an answer.

Ruling things out

The Survey keeps this list because operators who do not have it in front of them report the same eight things every year. In descending order of how often they turn out to be the answer: handling noise, the fridge, the boiler, a radiator or pipe expanding, traffic, a phone or laptop fan, a neighbour’s television through a wall, birds, and the operator’s own breathing — which is much louder and much more structured than anybody expects.

An incident record that does not say what was ruled out is not a record of anything. That applies here more than anywhere else on the bench, because this instrument will always find you something.

Ghost Detector Lab is an entertainment site operated as the public terminal of the Thinplace Survey. Nothing here detects the paranormal. See the disclaimer, the privacy notice, and the method for every instrument on the bench.