30% off your first 3 months

Get the deal

Getting a voice assistant to understand you in a noisy room

by DuckyHelperNov 12, 20264 min read
AI on your computer

TL;DR:

  • What matters is how loud your voice is compared with the background. Getting the microphone closer raises that ratio the most.

  • Short, clear questions give noise less room to cause a mistake.

  • In shared spaces, a headset or earbud microphone lets you speak softly and still be understood.

You are studying in a busy library and want to ask a quick question out loud. Keyboards clack, someone is talking two tables away, and you do not want to talk loudly enough to be heard across the room. You say it quietly, and the assistant answers a question you did not ask. This is a signal problem, and signals can be improved.

The short answer

A microphone picks up everything in the air, and the assistant has to find your voice within it. The measure that matters is the signal to noise ratio: how strong your voice is compared with everything else. You can raise it by making your voice louder at the microphone, which means moving the microphone closer or speaking up, or by lowering the noise. Closer is almost always the cheapest.

What happens step by step

  1. Sound from everywhere reaches the microphone, your voice and the room mixed into a single signal.

  2. The software tries to separate speech from the rest, for example by suppressing steady sounds like fans.

  3. It turns what remains into text. If the voice is weak, noise bits can be mistaken for speech sounds.

  4. The text goes on to be answered. An error here becomes a wrong answer.

Step 2 handles steady noise well and varied noise, like other people talking, poorly. Voices in the background look a lot like the voice you want.

Walk-through: asking in a busy library

Let us put numbers on one thing you can control, distance. In open air, sound from a small source gets weaker the farther you go, in a simple pattern: double the distance and the sound pressure halves. So the pressure is proportional to 1 / distance.

Say your mouth is 60 cm from a laptop microphone, and you move to 15 cm by using a headset or holding the laptop closer.

  1. The distance shrinks by a factor of 60 / 15 = 4.

  2. The voice pressure at the microphone rises by about 4 times.

  3. In decibels, 20 × log10(4) ≈ 12 dB louder, which is a big improvement.

  4. The noise in the room is far away from you, so its level at the microphone barely changes. The voice gets 12 dB stronger against it.

Check the direction: 4 times closer gives 4 times the pressure, and 20 × log10(4) = 20 × 0.602 = 12.04. The figure is an idealized estimate. Real rooms have reflections, and microphones are not equally sensitive in all directions, but the lesson holds: closer wins by a large margin.

Microphone position

  • Headset or earbud microphone. Sits a few centimeters from your mouth. The best option for noisy shared places.

  • Laptop microphone. Usually a hole near the screen. Face it, and keep it unblocked by your hand or notebook.

  • Direction. Many microphones favor sound from the front. Speaking across it, rather than into it, loses some voice.

  • Away from the noise source. Sit with your back to a noisy aisle or fan rather than facing it, if your microphone favors the front.

Avoid covering the microphone with your hand, and avoid putting the laptop on soft cloth, which can muffle it and cause scratchy sound.

Short questions

Noise causes occasional errors, not total failure. A short question has fewer places to go wrong, and one error is easier to catch. A rambling question has many, and a single missing word can flip the meaning.

Harder in noise

Easier in noise

Can you maybe go through how, um, the whole process of photosynthesis works and why plants need light?

What is photosynthesis?

Tell me about that thing from the chapter with the numbers

Explain the quadratic formula

Same need, two ways to ask.

Ask the short version first, then follow with another short question. You get to check each answer before you build on it. If a word is wrong, see fixing speech recognition mistakes quickly.

What it can't do

  • It cannot reliably pick one speaker out of several people talking nearby.

  • It cannot recover a word that the noise completely covered.

  • It cannot make a quiet whisper sound clear if the microphone is far away.

If you are in a place where talking is not appropriate, such as a silent study floor, respect the rule and type or step out. Your assistant is not worth a library fine.

Privacy and control

In public spaces, people around you can hear what you say, and a sensitive question should wait. Also check when your assistant listens. DuckyHelper listens only while the pill is open and for that answer, which means the microphone is not open when you stop. That matters more in a room full of other people's conversations.

Try it yourself

Run a test in a noisy place: ask the same short question three times, first at arm's length, then at half the distance, then with a headset microphone if you have one. Notice at which distance the text comes out right. That is your working distance for noisy rooms.

Frequently asked questions

Does speaking louder help more than moving closer?

Moving closer usually does, because it raises your voice without raising the whole room's noise, and it keeps you polite to your neighbors.

Will noise cancelling headphones help the microphone?

They quiet what you hear. Whether they also clean up what the microphone picks up depends on the model.

Why does it work at home but not in class?

Rooms with many voices are the hardest case, because other speech resembles yours. Closer microphone and shorter questions help most.

Get started

DuckyHelper can
do more

It sees your screen and works in any app you use. Try it for 10 minutes, no card.

Get it for Mac, Windows or Linux

Share

Questions and jobs

Do anything

Ask it a question or hand it a job, in any app.

Doing jobs for you is in early access on Mac. Asking works on Mac, Windows and Linux.