Dictation and Background Noise: Where and How to Dictate

Background noise makes dictation less accurate because the recognizer has to pick your words out of everything else it hears. The most reliable defense is a microphone close to your mouth, a room with little competing sound, and a steady, normal voice. Software settings matter less than those three things. This guide sets out what the evidence shows for noise, echo, open-plan offices, cars and public places, and what professionals do about the same problem with speech privacy masks and dictation rooms. For the basics of dictating at all, start with what dictation is. If you are asking a narrower question, see whether you can dictate in a noisy room.

Noise lowers accuracy because the recognizer cannot separate your voice from the room

A recognizer turns sound into words by matching the signal against speech patterns. Any other sound in the same signal competes with your voice, and the match gets worse as the competition grows. A 1996 study on recognition in noise put it directly: the performance of a speech recognition system degrades rapidly in the presence of ambient noise ¹. The authors modeled the Lombard effect, the way speakers change their voices when they talk in noise, and tried to compensate for it.

The ratio of your voice to the noise around it is one of the factors that matters most. In an oral-history archive with noisy, elderly and accented speech, signal-to-noise ratio and syllable rate were the two dominant factors explaining the overall word error rate ². You cannot change the archive, but you can change the ratio at your own microphone.

Dragon's own guidance treats noise as a basic accuracy issue. Its dictation guidelines tell users to reduce background noise, because Dragon hears and transcribes everything ³. The same page warns that, unless you tell Dragon to stop listening, it records everything it hears, including a phone conversation or a coworker ⁴. Other dictation tools may behave differently, but the principle is the same: whatever reaches the microphone is input.

A close microphone is the strongest defense against noise

Distance matters more than most people expect. In a study that re-recorded read speech with a distant microphone, recognition performance was dramatically reduced compared with close-talking speech ⁵. The same research group reported a 3.7 percent word error rate for close-talking read speech recorded in a studio ⁶. With a single distant ceiling microphone in an apartment, using real recordings, the error rate was 12.0 percent ⁷.

The two conditions differ in room and speech, so the ratio is only a rough guide. Divided, the distant figure is about 3.2 times the close-talk figure ⁸. The chart below shows the two error rates side by side.

Physics explains part of the gap. Under ideal free-field conditions, moving a microphone from 1 inch to 12 inches from the source lowers the level by about 21.6 dB ⁹. That is an idealized estimate for scale, not a measurement of your laptop, but it shows why a built-in laptop microphone a foot away is at a disadvantage against a headset at your cheek.

A useful way to think about this is that the microphone decides what the recognizer hears first. Once a sound is captured, software can only guess at what is missing. That is why a cheap headset close to the mouth often beats an expensive laptop microphone across the room, and why the first step in any noisy setting is to change the distance before you change any setting.

For a headset, Nuance advises keeping the microphone element about an inch from the side of the mouth, but not touching it ¹⁰. For a handheld microphone, it advises 1 to 3 inches from the mouth, not directly in front of it ¹¹. Apple advises that a headset microphone may help when Dictation is used in a noisy environment or in a room with excessive echo ¹². The choice of hardware is covered in the guide to dictation microphones, and the term is explained in the glossary entry for noise-canceling microphones.

Whispering is a different question. A quieter voice lowers the signal the microphone receives while the room noise stays the same, so the ratio that matters gets worse. See whether you can whisper to dictation software.

Hard, empty rooms add echo that hurts recognition

Echo, or reverberation, is the reflection of your voice off hard walls, glass and bare floors. The reflections arrive after the sound they copy, so each word blurs into the next. The Ravanelli paper names reverberation together with non-stationary noise as conditions where distant speech recognition still lacks robustness ¹³. Apple's advice on a headset microphone for rooms with excessive echo points the same way ¹².

In practice, a bare room with hard surfaces is likely to be worse than a furnished room, even if both are equally quiet. Curtains, carpet, upholstered chairs and bookshelves absorb reflections. A small, soft room is often a better dictation space than a large, empty office.

We did not find a current, open test that measures dictation error rates at several echo levels. Treat the echo advice as a direction, not a calibrated threshold. The guide to dictation mistakes covers the errors that echo and noise tend to produce.

Open-plan offices need a quiet voice, a close microphone and consideration for colleagues

Open-plan offices combine noise with the speech of other people, and that speech is widely seen as the main problem. A 2019 laboratory study of 55 people found a significant effect of speech intelligibility on performance, mental load and discomfort ¹⁴. The authors wrote that it seems accepted that speech noise in open-plan offices is the main source of discomfort for employees ¹⁵. Note that the study measured people doing a memory task, not dictation, so it shows the cost of nearby speech rather than a dictation error rate.

For your own dictation in a shared office, three things help. First, use a close microphone, so that your voice dominates the signal. Second, keep your speech at a normal conversational level and pace. Nuance tells Dragon users to speak clearly at a normal conversational tone and pace, and to dictate with a consistent speed and volume ¹⁶. Third, think about the people around you. Dictating loudly is the speech noise that the research describes, and it adds to the discomfort of everyone nearby.

Think about the spread of your speech as well. In an open office, a louder dictation carries across the room, and the people nearby pay the cost in concentration even when the recognizer gets your words right. A quieter, closer dictation is kinder to them and usually more accurate for you. If the room is too busy for dictation at all, a private booth, a meeting room or a close headset microphone is the practical alternative.

Shouting is not a fix. Speakers who talk in noise raise their voices and change how they speak, which is the Lombard effect the 1996 study models. A steady, ordinary voice at a close microphone is more accurate and more considerate than a loud one at a distant microphone.

For a fuller picture of the habit of dictating, see how to dictate and how to get used to dictation.

Speech privacy masks let professionals dictate without being overheard

A speech privacy mask is a hand-held or face-fitted enclosure with a microphone and sound-dampening material. The user speaks into it quietly, so that a room hears little or nothing. Wikipedia describes voice writing as a transcription method used for court reporting, medical transcription, CART and closed captioning ¹⁷. In that method, the reporter speaks directly into a stenomask or speech silencer, and the mask prevents the reporter from being heard during the testimony ¹⁸. A voice writing system consists of a stenomask, an external sound digitizer and speech recognition software ¹⁹.

The point of the mask is privacy as well as accuracy. Its microphone is close to the mouth, and its enclosure keeps the voice inside, which solves both problems at once. The glossary entry for voice writing explains the term, and the question on what a voice dictation mask is covers the device in more detail.

Manufacturers of these masks make strong noise claims, and we did not find an independent measurement that confirms them, so treat such figures as marketing claims. A mask is a sensible choice for a professional who must dictate in a room full of people. For most people at a desk, a headset microphone and a quiet room do the same job at much lower cost.

Hospitals and courts guard their speech spaces because the recording is the record

Hospitals and courts control their speech environments because the recording is the record. A transcript of a clinical note or a court record is only as good as the audio, and background sound can become text. Dragon's warning that it records everything it hears makes the same point for dictation software ⁴.

We could not find a published standard that defines a hospital dictation room. The term describes a space where clinicians record their notes. The evidence we found on such spaces is design practice, not measured results. A 2015 trade article reports that a design firm worked with radiology staff at Texas Children's to address concerns about noise control and lighting in a shared reading room ²⁰. Read it as an example of the design problem, not as a benchmark. The question of what a dictation room is has its own page: what a dictation room in a hospital is.

For clinicians who dictate notes, the practical lessons are the same as in any shared space: a close microphone, a quiet spot, and a habit of dictating one patient at a time. The medical dictation guide covers the workflows, and the glossary entry for digital dictation explains the systems that send recordings to transcriptionists.

Cars and public transport work for short notes, not long drafts

Car cabins are noisy in a steady way, and the noise changes with speed and road surface. A 2017 SAE paper on vehicle noise relates sentence error rate in cars to standard noise and vibration measures. Its conclusion is that AI%, an articulation-index measure, should be the preferred metric for relating cabin noise to speech recognition performance when the noise is steady ²¹. For a user, the paper's advice covers steady noise only. Road and wind noise change from moment to moment, and the abstract does not address them.

Steady and changing noise also differ in practice. A steady hum from a fan or an engine can be partly filtered out by the recognizer, but a voice in the background, a siren or a door that slams is a different signal, and it can be transcribed as words. Dragon's guidance is simpler: reduce background noise ³.

Public transport and street noise are harder to measure, and we did not find a comparable study for trains or buses. The same principles apply: a close microphone helps most, and crowds add speech that the recognizer may hear. Dictation in these places suits short messages and notes that you will check later. Long drafts are better written in a quieter room, then dictated or edited there.

Do not dictate while you are driving. Keep your attention on the road, and use a voice command only where the vehicle's own system is designed for it.

Check four things before you dictate in a noisy place

1. Put the microphone close. Keep a headset about an inch from the side of the mouth ¹⁰, not a laptop microphone across the desk.
2. Cut the steady noise. Close the door, turn off a fan or air conditioner if you can, and move away from machines. Steady noise is the kind a cabin or an open office produces, and it is the easiest to reduce.
3. Soften the echo. Choose a room with carpet, curtains or upholstery over a bare room with glass and hard floors.
4. Speak at a normal, steady level. Keep the same pace and volume throughout ¹⁶, and pause briefly before and after punctuation ²².

If you still see many errors after these steps, check the dictation mistakes guide for other causes, such as the wrong input device or a language setting. For how the recognizer turns sound into text, see how speech recognition works.

Where a dictation tool runs on your own device, the noise you capture is the only input it gets, so the setup advice above applies directly. Yap, which we make, is an on-device dictation app for Mac and Windows, and the same advice applies to it and to the built-in tools on your computer or phone. For the basics, see the dictation hub, the guide to dictation vs typing, and the glossary entry for dictation.