Microphones

You almost certainly don’t need to buy anything.

“The app can’t hear me properly” has a handful of causes, and most of them are routing rather than hardware. The wrong input was selected. Something invisible was sitting in front of it. The mic was fine and three feet away. All of those produce exactly the symptoms a bad microphone produces, all of them are free to fix, and none of them get better if you spend money first.

So this page starts with the fixes that cost nothing, then compares microphone types honestly — what each is genuinely good at and where each falls down. No brands, no rankings, no affiliate links. We haven’t tested specific models, so we’re not going to pretend otherwise.

Start with the free fixes

Costs nothing

Seven things to try before you open a shopping tab

In rough order of how often each one is the actual cause. The first four are about which device is being used; only the last two are about sound. That ordering is the point — people reach for acoustics when the problem is a dropdown.

01
Choose the input by name, not by luck
“Default” is not a microphone — it is a pointer that the operating system re-aims whenever the set of connected devices changes. Open your sound settings and select the input you actually mean, by its name. If you can’t tell which entry is which, unplug one thing and watch which name disappears.
02
Make the OS and the browser agree
A browser keeps its own microphone choice per site, stored separately from the system default. So you can set the system to your headset, and the page keeps recording from the laptop — because the site-level setting was made months ago and still wins. Check both. They are two switches for one thing, and only one of them is where you looked.
03
Restart the app after you swap devices
An audio stream that is already open holds on to the device it opened with. Plugging in a headset mid-session does not migrate it — nothing renegotiates halfway through a recording, by design, because switching devices under a live stream is how you lose audio. Plug in first, then open the app. If you plugged in after, close and reopen.
04
Turn off the virtual microphones
Tools like NVIDIA Broadcast, VoiceMeeter and Teams’ own audio processing install a fake input device that then presents itself as the sensible default. Everything downstream gets audio that has already been noise-gated by something tuned for a different purpose. Stacked suppression is the classic cause of “it hears me fine, then swallows every quiet sentence”. Disable them, or at least stop routing dictation through them.
05
Look at the level meter exactly once
You are checking for two failures, not chasing a number. Pinned to the top means clipping, and clipped audio is destroyed rather than quiet — no software recovers it. Barely moving means the wrong device is selected, or the mic is under a desk. Speak at your normal volume and aim for a meter that lives around three quarters and only occasionally touches the top.
06
Get the microphone out of the airflow
A laptop fan, a desk fan or an air-conditioning vent moving air across a microphone element produces broadband noise sitting right on top of your voice. It is the one kind of noise that gets worse the harder your machine works — which is to say, exactly while you are transcribing. Move the mic, or move the air.
07
Change the room before you change the mic
A rug, a curtain, a bookshelf, or simply not sitting in the middle of a bare-walled kitchen will do more for intelligibility than any purchase in the same afternoon. Hard parallel surfaces send a smeared copy of every word back into the mic a few milliseconds late, and that smearing is precisely what makes a transcriber guess.

Where the input picker lives

macOS — System Settings → Sound → Input. The list is your real device list; anything not here does not exist as far as any app is concerned.
Windows — Settings → System → Sound → Input, then “Choose a device for speaking or recording”. Check the per-app volume mixer too — a muted app-level input looks identical to a broken mic.
Any browser — The padlock or tune icon in the address bar → Microphone. This is per-site and overrides your system default, which is why it is the setting people miss.
iPhone and iPad — No per-app picker. The phone routes to whatever is connected, in its own priority order — so unplugging or unpairing is how you choose.

Two rules

Closer beats louder. Wired beats wireless.

If you remember nothing else from this page, remember these. Both of them beat almost any upgrade you could make, and both are free.

Rule one

Closer beats louder

Turning up the gain raises your voice and the fridge and the traffic and the fan by exactly the same amount. The ratio between you and the room — the only thing a transcriber actually cares about — does not move a decibel.

Moving closer raises only you. Halving the distance is a large, free improvement, and it is the reason a five-year-old pair of wired earbuds outperforms a good laptop mic: one is near your mouth and the other is near your keyboard.

This matters more in Waver than in most apps, because Waver compresses hard and applies makeup gain to make quiet speech usable. That helps you — and it lifts your room by the same amount. Distance is the one lever software can’t pull for you.

Rule two

Wired beats wireless

A wired mic is a device. It appears in the list by name, it is connected or it isn’t, and the audio path from the capsule to the app has nothing in it that has to negotiate anything.

A wireless headset is a negotiation. It sleeps, it reconnects, it changes its own audio format depending on what you asked it for, and it decides some of this without telling you. Every one of those is a way for the beginning of a sentence to go missing.

For listening, none of that matters. For dictation, all of it does — which is the whole of the next section.

The sixty-second version: record the same sentence three ways

Pick a sentence with a couple of proper nouns and some numbers in it — numbers are where marginal audio shows itself first, because “fifteen” and “fifty” differ by consonants that compression eats. Record it once on your laptop mic at your normal seating distance, once on whatever wired earbuds you own, once on your phone held near your mouth. Read all three transcripts side by side. That comparison will tell you more about your setup in one minute than any spec sheet, and it costs nothing.

Types, not brands

Six shapes of microphone, and what each one is actually like

Every one of these has a real drawback. If a guide gives you a category with no downside, it is selling you the category. The differences that matter for dictation are mostly about where the mic sits relative to your mouth and what else it can hear from there — not about the capsule inside it.

Laptop, tablet, phone

The built-in mic

The one you are not counting. Phones in particular are excellent for dictation for an unglamorous reason: the mic ends up a hand’s width from your mouth, which beats any amount of engineering happening an arm’s length away. Laptops are the weaker case — the array is in the lid or top edge, aimed at where your face usually is, and it hears the keyboard you are typing on as clearly as it hears you.

Good at: Always present, nothing to select, nothing to pair, no profile negotiation, no battery.
Bad at: On a laptop it is far away and shares a chassis with the fan and the keys. Blocked or muffled with the lid closed.

3.5 mm or USB-C

Wired earbuds with an inline mic

The most underrated option on this page and usually the cheapest thing in the drawer. The capsule dangles on the cable somewhere around your collarbone, which is roughly four times closer than a laptop lid — and closeness is the whole game. Plugging them in also makes the routing unambiguous, since a wired input appears as its own named device rather than as a mode of something else.

Good at: Close to your mouth, instant, no pairing, no battery, no codec switch. Costs nothing if you own a pair.
Bad at: The cable rubs on clothing and every rub is in the recording. Fixed position, so turning your head changes your level.

The call-centre shape

A wired headset with a boom

The most consistent audio you can get without thinking about it, because the boom holds the capsule at a fixed distance from the corner of your mouth no matter how you move. That consistency is worth more than raw quality: a transcriber copes far better with mediocre audio that never changes than with good audio whose level swings every time you lean back.

Good at: Same distance every second of every session. Rejects the room well. You hear playback at the same time.
Bad at: You are wearing it. On camera it reads as a support desk. Boom too close and every P becomes a thump.

Wired, on your shirt

A clip-on lavalier

Solves the same problem as a boom without anything on your head, which matters if you pace while you think or if you are on camera. It clips at roughly a hand’s width below your chin and stays there, so walking around the room doesn’t change your level the way it does with a desk mic.

Good at: Constant distance while you move. Nearly invisible. Comfortable for hours.
Bad at: Collar rustle and jacket zips are recorded at point-blank range. Cable has to go somewhere. You have to remember to clip it on.

Parked in front of you

A gooseneck or desk-arm mic

You put it where you want it once and it stays there, which suits long seated sessions where wearing anything gets old. The catch is mechanical rather than acoustic: a desk is a drum, and unless the mic is on a shock mount or a decoupled arm, every keystroke and coffee-cup landing travels up the stand and into the capsule as a thud.

Good at: Nothing on your body. Aim it once. Easy to mute by leaning away.
Bad at: Transmits desk knocks and typing through the structure. You have to stay in front of it, which you will forget to do.

The podcast-looking one

A USB condenser

The best raw capture of the six, and the one most likely to disappoint. A sensitive mic in an ordinary room is an accurate record of an ordinary room — the fridge, the street, the reflection off the window behind you. It rewards a treated space and punishes a bare one, and for dictation specifically it is almost never the thing that was limiting you.

Good at: Most headroom and detail. Often has a real mute switch and a headphone jack for latency-free monitoring.
Bad at: Hears the room faithfully, including the parts you had stopped noticing. Takes desk space. Needs a stand or arm to be worth it.

Or skip all that and pick by situation

What you’re doing	What to use
Dictating alone in a quiet room	Whatever is built in. Genuinely — spend the money on nothing.
Open-plan office or a café	Wired boom headset. You need the room rejected, not the mic improved.
Pacing around while you think	Your phone, held near your mouth, or a clip-on lavalier.
On camera as well as on mic	Clip-on lavalier. Same benefit as a boom, without wearing a boom.
Long seated sessions, hours a day	Desk arm or gooseneck. Comfort is the spec that decides this one.
A meeting with people in the room	Something in the middle of the table. Distance is the enemy, and it multiplies per person.
Laptop closed on a stand	Any external input, chosen by name. See the clamshell section below.

The one to avoid

Bluetooth is the worst thing you can dictate into

Not because wireless is inherently bad, but because of what a headset has to do the moment an app asks it for a microphone. Four separate things go wrong, and they all go wrong at the start of your sentence.

It changes format the instant you record

Playing music, a headset runs a one-way stereo mode with bandwidth to spare. Ask for the mic and it has to switch to the two-way call mode, which splits one small radio budget between both directions. Historically that meant narrowband audio; the wideband option is better and still a fraction of what the same buds carry for music. You can hear the moment it happens — the music dulls. That dulled audio is what the transcriber receives.

The switch is not instantaneous, and you talk through it

Negotiating the change takes a moment. You press record and start speaking, because from where you are sitting the recording has started. So the transcript opens mid-word: you said “Reschedule the Tuesday review”, the note begins at “schedule the Tuesday review”, and it reads like a transcription error rather than the audio never having existed.

Idle buds have to wake up first

Wireless earbuds power down aggressively to protect battery life. If they have been sitting in your ears doing nothing, the first thing your record button triggers is a reconnection, not a recording. This is the same failure as above with a longer fuse, and it is worst exactly when you are capturing a thought quickly — which is the entire reason for hold-to-talk.

The audio is already compressed before any app sees it

Whatever the link does to your voice happens first and cannot be undone downstream. The first things to go are the high-frequency details — the difference between an S and an F, the burst at the front of a T — which are exactly the cues that separate “fifteen” from “fifty” and one name from another. Speaking more loudly does not bring them back; they were discarded on the way.

If wireless is the only option

Press, breathe, then speak. Half a second of silence after you start covers the negotiation and costs you nothing — Auto Cleanup drops the pause anyway.
Split output from input. Keep the buds for listening and set your input to a wired or built-in mic. With no app asking the headset for its microphone, it never switches out of the good mode. Waver captures with echo cancellation on, so audio coming from your speakers is handled.
Check what your gear supports. Newer wireless standards carry usable bandwidth in both directions without the old mode flip, but both ends have to support it, and plenty of current hardware doesn’t. Worth confirming rather than assuming.

The trap nobody suspects

A closed laptop still has a microphone. That’s the problem.

Lid shut, machine on a stand, external monitor and keyboard — the setup half of us work in. On most laptops the mic array lives in the lid or along the top edge, so closing the lid presses it against the keyboard deck or hides it behind the hinge.

The reason this goes undiagnosed for months is that it doesn’t look like a failure. Nothing errors. The level meter still moves. You still get a transcript. It is just muffled, and the room is now louder relative to you than it was — so the accuracy degrades gradually instead of breaking, and you conclude the transcription “got worse”.

Docks reshuffle the device list

Connecting a dock or a monitor can add inputs — some displays and webcams have their own mic — and the operating system may quietly re-aim “default” at one of them. A monitor mic is across the desk from you, which by rule one is the worst position available. Choose the input by name after docking, not before.

Restarting matters more here

Docked setups are exactly the case where you plug things in mid-session. A stream already running keeps its old device, so the input you just selected does not take effect until the app reopens. This is the single most common reason a correct setting appears to do nothing.

Some machines disable the internal mic in clamshell, some don’t

Behaviour varies by manufacturer and model, and neither outcome is announced. Either way you should not be depending on it: if the lid is shut, decide on an external input deliberately rather than discovering afterwards which of the two your laptop does.

Watch where the fan exhaust points

A closed laptop on a stand runs warmer and works its fans harder, and a vent aimed at a desk mic is a continuous wash of broadband noise that peaks precisely when you are transcribing something long. Move one or the other.

Our half of the job

What Waver does so the microphone matters less

None of this replaces sitting closer. It is the part we can do from our side of the cable, and it is why the built-in mic is usually enough.

The capture chain is tuned for a voice, not for music

Audio comes in as mono at 48 kHz, through a high-pass filter at 80 Hz — below the bottom of a speaking voice, so it removes mains hum, fan rumble and desk thumps without touching anything you said. After that it is compressed hard and gain is applied, which is what lets a genuinely quiet voice reach the transcriber at a usable level.

Browser noise suppression is deliberately switched off

This is a real trade-off, made on purpose. Echo cancellation and automatic gain stay on; the browser’s noise suppression does not, because it is aggressive enough to erase a whisper completely — and the transcription model handles background noise better than a gate does. The cost is that a noisy room sounds noisy in your recording. We would rather keep the quiet sentence and the noise than lose both.

Your Personal Dictionary is sent with the audio

Names, jargon, product terms and the words you have corrected before travel alongside the recording rather than being fixed up afterwards. That is the difference between a transcriber guessing phonetically at a marginal signal and one that has already been told which words are plausible. It is the single biggest accuracy win available to someone on an ordinary mic.

Auto Cleanup has four levels, and the original is always recoverable

None, Light, Medium or High — because a rough mic and a rough room produce more filler and false starts, and how much of that you want removed is a preference, not a setting we should decide. Whatever it does, “what I actually said” restores the raw transcript, so cleanup can never quietly lose you a word you needed.

Hold-to-talk, so recording starts before you do

The Flow Bar records while you hold, in-app on web and mobile and system-wide on the desktop build. Holding a key is a deliberate act with an obvious start, which is exactly the discipline that stops the first word going missing — and the reason we recommend a breath before you speak on any wireless setup.

Honest exceptions

When buying something is actually the right call

There are three. Notice that all three are about the situation you are recording in rather than about audio quality in the abstract — which is why “get the better mic” is the wrong question and “get the mic closer, in this specific room” is the right one.

01
You don’t control the room
Open-plan office, shared flat, café, a house with other people in it. No amount of processing separates you from a conversation happening six feet away, because to a microphone that conversation is speech too. The fix is physical: get the capsule close to your mouth and let distance attenuate everything else. A boom headset or a clip-on, not a better desk mic.
02
You are recording other people, not only yourself
The moment there are four voices around a table, the person furthest away sets your accuracy, and they are the one you cannot ask to lean in. Something in the middle of the table beats anything in front of you. Note that this only applies to people physically present — Waver’s meeting recorder captures system audio as well as your mic, so remote participants arrive clean regardless of what you own.
03
You do this for hours a day
At that point the deciding spec is comfort, not frequency response. Whatever you can wear or sit in front of all afternoon without noticing is the correct answer, and it is worth paying for the version that doesn’t pinch. Nobody has ever abandoned a dictation habit because their microphone was insufficiently detailed.

And the one thing not to buy first

A large, sensitive studio-style mic for an untreated room. It will faithfully reproduce the room, which is the opposite of the goal — you wanted more of your voice relative to everything else, and you bought a device that hears everything else better too. If a room is the problem, soften the room. If distance is the problem, close the distance. Buy the sensitive mic when neither of those is true any more.

Questions

The ones that come up every time

Do I need to buy a microphone to use Waver?

No. Waver is built to work with the mic already in your laptop or phone — that is the audio it was tuned against. If your transcripts are poor, work down the free fix list first: the wrong input being selected and the selected input being bad look identical from where you are sitting, and only one of them costs anything to rule out.

Where do I choose which microphone Waver uses?

In your operating system. Waver records from whatever your system has selected as the input; there is no separate device picker inside Waver. That is deliberate — a second picker gives you two settings that can disagree, and then the bug is not that the audio is bad but that you cannot tell which switch is in charge.

Why does the first word of my recording keep disappearing?

Two usual causes. On a wireless headset, the link has to switch into its two-way call mode when an app asks for the mic, and that negotiation takes a moment — you start talking during it. On hold-to-talk, it is simpler: the key is registering fractionally after you have already begun. Press, take a breath, then speak. You will not notice the pause, and the first word survives.

Are wireless earbuds fine for dictation?

Workable, but they are the worst common choice, and the reasons are structural rather than a matter of price. Bandwidth in a headset’s two-way call mode is a fraction of what the same buds use for music, the switch into that mode costs you the start of the sentence, and the audio is compressed before your app ever sees it. If you can plug something in instead, do.

Should I turn on noise cancellation on my headset or in my OS?

Usually not for dictation. Waver deliberately leaves the browser’s own noise suppression off, because it is aggressive enough to erase a quiet sentence entirely and the transcription model copes with background noise better than a gate does. Adding a second layer of suppression upstream is what produces the failure where normal speech transcribes fine and anything softer becomes silence.

My level meter moves, so why is the transcript still wrong?

Level and clarity are different things. The meter tells you something is arriving, not that your voice is louder than the room, and gain raises you and the fridge by exactly the same amount. Halving your distance to the mic is the only free move that raises one without the other. If the words being missed are names, jargon or product terms, that is a different problem — add them to your Personal Dictionary, which is sent along with the audio.

Does the meeting recorder need a better microphone?

For remote participants, no — Waver captures system audio alongside your mic, so people dialling in arrive as clean digital audio and never touch a microphone at all. The mic only decides how well the people physically in the room with you come out. If that is a regular thing, put something in the middle of the table rather than upgrading the mic in front of you.

Does mic quality matter more in some languages?

Clean audio helps everywhere. It tends to matter most when you are switching between scripts mid-sentence, or using names the model cannot infer from surrounding context. Waver transcribes 120+ languages in their own script and auto-detects which one you are speaking, but no amount of language coverage recovers a word the microphone never captured. Names and jargon are what the Personal Dictionary is for.

Will a USB mic work on Windows and Mac?

If your operating system lists it as an input, Waver can record from it — on macOS, Windows, iPhone and iPad. Waver does not maintain its own device drivers or hardware list, which is exactly why there is nothing to check for compatibility.

Is any of this worth doing before I have tried Waver?

No. Record something first on the mic you already have. The free tier includes unlimited transcription with no card, so the cheapest possible test of whether your audio is good enough is to find out.

Still not getting clean transcripts after all of this? Tell us what you tried — a description of your setup and one bad transcript is usually enough for us to spot which of the seven it was.

Stop typing what you could have just said.

Free forever to start. Unlimited transcription, no card, no ads.