# Your room and background noise

Echo, fans, open-plan offices and the colleague talking behind you — what actually helps, in order of effort

Blabb transcribes what it hears. It has no way of knowing which voice in the room is yours, and no
way of knowing that the hum under your sentence is the air conditioning rather than something you
said. Everything below is about giving it less to sort through.

## What the room does to your voice

Hard surfaces are the usual culprit. A desk, a window and two bare walls send a copy of every word
back into the microphone a few milliseconds late, so the model hears each syllable twice, slightly
smeared. That is why the same person sounds crisper on a headset in a bare room than on a desk
microphone in the same room — the headset is close enough that your direct voice drowns out the
reflections.

Steady noise is a different problem. A fan, a fridge, an air conditioner or a projector produces a
constant sound that never stops for you to pause in. Blabb applies a high-pass filter at 80 Hz
before it adjusts the gain, which removes the low rumble underneath that sort of noise. It does not
remove the rest of it, and it does nothing at all about a conversation two desks away, which sits
in exactly the same frequency range as your own voice.

Sharp, brief noises land in the transcript as odd single words. A mechanical keyboard is the
common one: if you type while you dictate, and the microphone is the laptop's own array a few
centimetres from the keys, some of that will come through.

## A second voice

This is the honest limit. Blabb does not separate speakers and does not pick yours out of several.
If a colleague is talking near you and the microphone can hear them, their words can end up in your
text alongside yours.

The fix is physical, not a setting. Get the microphone close enough to your mouth that you are
dramatically louder than they are — a boom mic at 5 cm makes a voice at three metres almost
irrelevant. There is no switch in the app that does the equivalent.

## Fixes, in order of effort

| Effort | What to do | What it fixes |
| --- | --- | --- |
| None | Move the microphone closer to your mouth | Echo, room noise, other voices — all of it at once |
| None | Stop typing while you speak | Keyboard clatter in the transcript |
| A minute | Turn off the fan or close the window while you dictate a long piece | Steady hum |
| A minute | Turn the laptop so the vents are not blowing towards the microphone | Fan noise on a built-in array |
| An afternoon | Put something soft in the room — a rug, curtains, a bookshelf, anything not flat and hard | Echo and smearing |
| A move | Dictate somewhere else: a meeting room, a corner, the car | Open-plan noise and other voices |

The pattern is worth naming, because it saves you time later. Almost every improvement on this list
is a version of the same thing — increase the gap between how loud you are and how loud everything
else is. Getting closer does that for free. Everything else is working around not being able to.

> [!TIP]
> If you have to dictate in a noisy place, dictate in shorter pieces. A segment ends at real
> silence, so in a room with no silences you are relying on your own pauses to give Blabb a clean
> break.

## What to check before blaming the room

If accuracy is poor everywhere, including in a quiet room, it is probably not the room. Work
through [choosing and placing your microphone](/docs/accuracy/microphone-setup/) first — a laptop
array at arm's length produces the same mushy result in a library as it does in a café.

## See also

- [Choosing and placing your microphone](/docs/accuracy/microphone-setup/) — the fix for most of
  what is on this page.
- [How to speak so Blabb hears you](/docs/accuracy/how-to-speak/) — the half of this that is you
  rather than the room.