By Denys MedvedievTroubleshooting
Mic works, dictation doesn't
Your mic test shows the levels moving, but voice typing produces no text. Almost always it's the speech layer, not the mic: an online-speech toggle that's off, a language mismatch, or the wrong field in focus. The fixes below are per-OS and conservative.
Last updated: June 2026
When the microphone works but dictation doesn't, the mic is fine and the speech layer isn't. On Windows, turn on online speech recognition and confirm the input language. On Mac, check the dictation language and microphone permission. A dedicated offline tool sidesteps the whole online-speech handshake that usually causes this.
The first time this happened to me, I spent twenty minutes blaming a perfectly good microphone. The level meter was bouncing, the headset light was on, the test recording played back fine — and dictation still gave me a blinking cursor and nothing else. The mic was never the problem. The problem was the layer that sits between the mic and the text, and that layer fails quietly.
This is the most common confusion in dictation: a mic test only proves audio reaches the operating system. Dictation needs a second thing — a speech-recognition service that's turned on, set to the language you're speaking, and pointed at the same mic, with your cursor in a field that accepts text. Any one of those being off gives you exactly what you're seeing: a working mic and no words.
Here's the part most "fix it" pages skip. Your microphone and your dictation are two separate systems. The mic captures sound. The speech recognizer turns that sound into text. A mic test exercises the first one and tells you nothing about the second.
So the question isn't "is my mic broken" — you've already answered that. The question is "which link in the speech chain is down," and the answer is usually one of five: the online-speech service is off, the dictation language doesn't match what you're saying, the field isn't really focused, the wrong mic is selected for the speech service, or the on-device language pack hasn't downloaded. I'll walk the fast fix, the permanent one, and the per-OS repair if you'd rather keep the built-in tool.
Why the mic passes but no text appears
A working mic and working dictation are not the same test. When you open Sound settings and watch the input level move, you've confirmed audio is reaching the operating system. Dictation sits one layer up: it takes that audio and runs it through a speech-recognition service. If that service is off, misconfigured, or listening to a different device, the meter still moves and no text appears. That gap is the whole bug.
On Windows 11, voice typing — the Windows logo key + H bar — converts speech using Microsoft's online speech recognition, which runs in the cloud and needs an internet connection. Microsoft's own support page is explicit: voice typing requires a working microphone, a text box for your cursor, and an internet connection. If the "online speech recognition" privacy setting is off, or you're offline, the bar opens, the mic light comes on, and nothing lands. This is the single most common version of "mic works, dictation doesn't" on Windows.
On macOS, the failure modes are different but rhyme. Dictation has to be enabled in Keyboard settings, set to the language you're actually speaking, and granted microphone access. On Apple Silicon, many languages run on-device after a one-time download — and if that download hasn't finished, dictation can sit there doing nothing. Apple's troubleshooting guide also flags the obvious-but-common ones: wrong input source selected, wrong language for your speech, an obstructed mic. None of those show up in a level-meter test, which is exactly why the meter fooled you.
The fast fix that clears it for most people
Most cases come down to two or three checks per operating system. Run them in order and stop when text appears. They're drawn from Microsoft and Apple's own support pages, not from guesswork, because a wrong instruction here wastes an afternoon you don't have.
On Windows, do these three. First, turn on online speech recognition: Settings > Privacy and security > Speech, switch it on. Second, confirm you're connected to the internet — voice typing won't transcribe offline. Third, check the input language: press the Windows logo key + Spacebar and pick the language you're speaking. Then put your cursor in a real text box, press Windows logo key + H, and talk. If it still does nothing, confirm the right mic is selected under Settings > System > Sound > Input, and that mic access is on under Settings > Privacy and security > Microphone.
On Mac, do these three. First, make sure Dictation is enabled: System Settings > Keyboard > Dictation, turn it on. Second, set the dictation language to match what you're speaking — under that same Dictation panel, edit the language list. Third, confirm the right input is chosen under System Settings > Sound > Input, and that the app you're in has microphone permission. Click into a text field, trigger Dictation with its keyboard shortcut, and speak. The overlay below is from a different kind of tool — a dedicated dictation app that skips the online-speech service entirely — but the idea is the same: a small indicator that confirms it's actually listening, so you're never guessing.
The permanent fix: dictation that never hits this
If you'd rather not re-fight an online-speech toggle every few weeks, the durable move is a dictation tool that doesn't depend on the operating system's speech service at all. Whisper runs its own engine locally, uses your mic directly, and pastes at the cursor. There's no cloud handshake to time out and no privacy toggle to fight — so a working mic genuinely means working dictation. You need a Mac on Apple Silicon or a Windows 10-or-newer PC, a working microphone, and a couple of minutes. The local pipeline is free for any signed-in account, with no payment method at sign-up.
Step 1 — Install Whisper and sign in.
Download from the download page, install, and create a free account. No card. The local transcription pipeline opens right away.
You'll know it worked when the app's tray icon appears and the setup offers to pick a model.
Step 2 — Pick a transcription path and let a model download.
The app doesn't choose for you. You get three: Cloud (OpenAI, bring your own key), Local Parakeet, or Local Whisper. For a fully offline fix, start local — more on which one two sections down.
You'll know it worked when a model finishes downloading and shows as ready.
Step 3 — Confirm your hotkey and mic.
Windows defaults to Ctrl+Space, Mac to Command+Option held as push-to-talk. Point it at the same mic that passed your test. On Mac, grant the Accessibility permission when prompted, or the paste-at-cursor can't reach other apps.
You'll know it worked when a test recording pastes into any text field.
Step 4 — Put your cursor anywhere and talk.
Click into any text box, hold the hotkey, say a sentence, release. The transcript appears where the cursor is, with no online-speech service in the loop.
You'll know it worked when your spoken sentence is sitting in the field as text.
The slow part is the one-time model download, not the setup. Once a local model is in place, the speech engine lives on your machine — there's no toggle in Privacy settings that can quietly switch it off, and no internet outage that makes it go silent. That's the structural reason this class of bug doesn't recur here.
If you'd rather repair the built-in one
Keeping the OS dictation is a perfectly good choice, and the deeper fixes are worth knowing. On Windows, the usual culprit is online speech recognition being off after an update or a privacy sweep: Settings > Privacy and security > Speech, turn it back on. If it's on and still failing, Microsoft's order is to recheck the input device at Settings > System > Sound > Input, confirm microphone access at Settings > Privacy and security > Microphone, and verify the language with Windows logo key + Spacebar. Because Win+H routes through the cloud, also confirm you're online — an otherwise-correct setup produces no text the moment the connection drops.
On Mac, start at System Settings > Keyboard > Dictation and confirm it's enabled. The two quiet failures are language and download. If the dictation language doesn't match the language you're speaking, you'll get nothing or nonsense — edit the language list in that same panel. And on Apple Silicon, many languages process on-device only after the speech model finishes downloading, so give it time and a connection the first time. Then check the input under System Settings > Sound > Input and make sure the app you're dictating into has microphone permission. Apple's own list adds the boring-but-real ones: an obstructed mic, the wrong input source, too much background noise.
Two cross-platform traps catch people regardless of OS. First, focus: dictation types into the field that has the cursor, so if a notification stole focus or you're in a control that doesn't accept text, the words have nowhere to go. Click directly into the text box and try again. Second, the wrong mic: if the speech service is bound to your laptop's built-in mic but you're talking into a USB headset (or the reverse), the meter you tested can move while the recognizer hears silence. Make the device you're speaking into the selected input, not just a default that happens to pass a test.
Local or cloud: which mode to pick
If you move to a dedicated tool, the choice is which engine runs your dictation. The app makes you pick rather than guessing for you, and the right pick depends on language and how much you care about staying offline. For the "mic works, dictation doesn't" crowd specifically, local is the satisfying answer, because the entire failure you just hit comes from an online-speech service — and local removes that service from the picture.
Here's how the three paths differ:
- Local Parakeet — NVIDIA's TDT engine, around 600 MB, and the fastest local option — 5 to 10 times faster than Whisper on CPU. Covers English plus 24 other European languages, 25 in total. No translate-to-English. If you dictate in English or another European language, this is the quick, fully offline pick.
- Local Whisper — slower than Parakeet on the same machine, but the multilingual builds cover 99 languages and can translate to English. The English-only builds are English-only, not 99. Pick this for Chinese, Japanese, Korean, or any translation work Parakeet can't do. The default English model is around 480 MB.
- Cloud (OpenAI, BYOK) — best accuracy and web access, using your own OpenAI key billed straight by OpenAI. Transcription runs on gpt-4o-mini-transcribe by default. It needs internet, so it's the one path that leaves your machine — the same dependency that started this whole problem. The Cloud surface is part of Whisper Pro.
The boring truth is that for everyday dictation, local is plenty, and it's the mode that structurally avoids what you just debugged. Both local engines run fully on your machine with nothing sent to a server. Cloud earns its place when you want top-tier accuracy on a hard recording or you need the model to pull a fact off the web mid-sentence. If the trigger for reading this was "the online thing stopped working," local is the answer that doesn't have an online thing to stop.
Cleaner text once dictation works
Once words are landing, the next annoyance is that raw dictation comes out as a run-on. You say "okay so reply to the school email confirm Thursday and remind me to send the form," and that's the unpunctuated wall any speech engine hands you. Cleaning it up is where the tools differ.
Windows Voice Typing adds punctuation as you speak, and macOS Dictation handles basic punctuation when you say "comma" or "period." For heavier cleanup — stripping the "ums," fixing the run-ons, turning a spoken paragraph into something you'd actually send — Whisper can run an AI pass. Say the activation phrase "Hey whisper" and the text gets enhanced before it lands. On a local model that runs through Ollama; in cloud mode it's gpt-5-mini by default.
okay so reply to the school email confirm thursday and remind me to send the form um before friday
Okay, so reply to the school email, confirm Thursday, and remind me to send the form before Friday.
The cleanup pass is also the moment to fix the small misfires every recognizer makes — a name it spelled phonetically, a number it heard as a word. A local model handles this on your machine; cloud does it with more headroom on harder audio. Either way, the point is the same: get the words down by voice, let the pass tidy them, and stop retyping what you already said.
That speak-then-clean flow pays off everywhere, not just in one app — you can type faster across every program with your voice once the mic-to-text path is solid, so a long message becomes a few spoken sentences instead of a paragraph you peck out.
When the built-in is enough
Sometimes this was a one-off, and pretending you need a new app would be dishonest. If flipping the online-speech toggle back on, or matching the dictation language to what you're speaking, fixed it — and the built-in tool now does what you need — you're done. Don't install anything. The free dictation already on your machine is fine for short bursts.
On Windows, if voice typing works again after you turned online speech recognition back on, and you're usually online anyway, Windows logo key + H is right there and free. On Mac, if Dictation came back once the language and permission were sorted, System Settings > Keyboard > Dictation is all you need, and on Apple Silicon it can run on-device once the language has downloaded. For a quick reply or a two-line note, that's the correct tool, and it costs nothing.
Reach for a dedicated, offline tool when the built-in keeps relapsing: the toggle that won't stay on, dictation that dies whenever the connection wobbles, long-form work where the online cap or the cloud round-trip gets in the way, or wanting one hotkey that behaves the same everywhere. Below that bar, use what's free. I'm not going to tell you to install an app to fix a setting you can flip in ten seconds.
If the failure was Windows-specific and you want the full per-version walkthrough, the deeper guide on voice to text not working on Windows covers the update-related breakages and the modern replacements in more detail.
The mic was innocent the whole time. It usually is. Dictation is two systems wearing one trench coat — a mic that captures sound and a recognizer that turns it into text — and the level meter only ever vouches for the first one. Flip the right toggle, match the language, click into the right box, and the words show up. I wrote most of this with a tool that skips the recognizer-handshake entirely, which is one way to make sure the trench coat never falls open mid-sentence.
Get dictation that just listens
Point it at the mic that already passed your test, hold the hotkey, and talk. The text lands at your cursor — no online-speech service in the loop to misconfigure.
Free local mode for any signed-in account. No card required to start.
