Habits from the room around you · habit 37 of 40

Saying it out loud

Long strings are hard to check in silence, so people say them: under the breath, half aloud, or straight into a microphone. It is the oldest error correcting trick there is and it genuinely works.

What it does everywhere else

Speaking a string brings a second sense in to check the first. Anybody who has read a serial number down a phone line knows the rhythm of it, letters grouped in fours, numbers called separately. Warehouses, pharmacies and call centres run a version of the same drill, because the mouth catches what the eye slides over. Dictation extends the trick: for many people speech is faster and more accurate than thumbs on glass.

What it does on this one

The string here is an onion address, and an address said aloud reaches the whole room in one go. Everybody present hears it. Any device waiting for a wake word hears the words around it, and a false trigger starts a recording exactly like a real one. If the words went into a dictation box, then on many setups they left the device to be turned into text somewhere else.

The habit

Reading an address aloud, or dictating it, so that the characters come out in the right order.

Why it is automatic

The training is older than the technology involved. Telephones taught a whole century to spell things out loud, and every workplace that handles reference numbers still does it. Speech forces the reader to attend to each character in sequence, which is precisely what silent reading refuses to do with a random string.

Random strings are the hard case. Ordinary words carry redundancy, so a missing letter still reads. A string with no meaning has none of that, which is why an eye passes over a wrong character several times without registering it. The mouth cannot skip.

For some readers speech is not a shortcut but the main way of putting text into a machine, for reasons of hands, sight or plain preference. That deserves saying, because advice written as though everybody types is useless to them. The point is not that speech is a bad input. It is that this one string travels further as sound than anywhere else.

What it actually does here

A room is rarely empty. There are other people in it, or through a wall, or on the far end of a call somebody forgot was connected. A voice does not stay at the desk. It carries into a hallway, out of an open window, and into the memory of anybody who heard something odd enough to remember, which a long nonsense string very much is.

The other listeners arrived recently and quietly: a smart speaker, a phone on the table, a television, a watch, a car. Wake word detection runs constantly by design and it is not perfect. A false trigger begins a recording of whatever follows it, and on most consumer setups that audio goes to a server to be interpreted. It is the standing configuration of the room rather than a choice made this evening.

Dictation works the same way and more so. By default on many phones and desktops, speech typed into a text box is processed off the device rather than on it. Some systems offer an on device mode and some do not, and the setting is made once at setup and never seen again. A mars market address dictated into a field has been read out to at least one system that had no reason to hold it.

The last property matters most. Text can be deleted and a clipboard can be cleared. A spoken string cannot be unspoken. Once it has left as sound it is in a room, possibly in a recording and possibly in a person head, and none of those three has a delete key.

Why reading aloud works at all

The benefit is worth keeping, because the accuracy problem it solves is real. What speech provides is serialisation: one character at a time, in order, at a pace the eye will not adopt by itself. Anything else that imposes the same discipline gives most of the same result without any sound.

The silent versions are easy. Run a fingertip under the line so the eye cannot jump. Cover the rest with a card and move it down. Compare in groups of four, because four is something a person can hold in mind and a long line is not. Better still, let the clipboard carry the string and use your eyes on the first group and the last, which is where an altered copy usually shows.

Everything in the room that is listening

The useful exercise is not to distrust devices but to be able to name them. Most people name the smart speaker and stop there.

The phone on the desk, face down, which does not stop the microphone.
A watch or a fitness band, which usually has one and usually has a wake word.
Headphones and earbuds, including a pair sitting in an open case.
A television or a remote control with voice search on it.
A car on the drive, if a phone is paired to it and a window is open.

What to do instead

The aim is to keep the accuracy and lose the broadcast, which is mostly a matter of swapping one checking method for another.

Check silently: finger under the line, compare in groups of four, and look hardest at the beginning and the end of the string.
If you do voice something, know which devices in the room have a microphone, and remember that a whisper is still audio to one on the same table.
Turn dictation off in the browser you use for this, or switch the system to on device speech recognition if it offers one.
Never read an address into a call, a voice note or a video, however well you know the other person. The recording outlives the conversation, which is the theme of the section on other people.
If speech is how you write, draw the line by task: ordinary text by voice, this one string by clipboard and eye.

How to notice you did it anyway

Spoken things leave traces in places nobody thinks of as storage, and most of them are viewable by the account holder.

The voice or assistant activity page on each account you use, which usually lists recordings by date and often plays them back.
Voice notes and half sent recordings in messaging apps, where something recorded by accident is easy to miss.
Meeting recordings and automatic transcripts on any work system, which is one more reason for the work machine section.
Smart speaker history, and the history on a television or a car with voice search enabled.
Other people in the house repeating a string back at you, which is the clearest sign it travelled.

What readers ask about this habit

Is muttering under my breath the same thing?

For a person across the room, no. For a microphone on the same desk, largely yes, since it is close and the recording gain does the rest. The distance to the nearest microphone matters more than how loudly you speak.

Does turning off the wake word solve it?

It removes the largest and most casual listener, and it is worth doing on a device that sits in the room you work in. It does nothing about people, about a call still connected, or about dictation, which is a separate feature with its own setting.

I dictate because typing hurts. What should I do?

Keep dictating for everything else and treat the address as the exception, since it is one short task rather than a workflow. If the system offers on device recognition, turning it on covers the rest. Where an address is needed repeatedly, the page on paper notes is usually a better fit than voice.