Transcription suddenly got worse or feels less accurate

Mac

1. Force quit and restart Flow. Press Command + Q. If Flow doesn't close, click Apple menu → Force Quit, select Wispr Flow, and click Force Quit. Reopen the app.

2. Check microphone permissions. Apple menu → System Settings → Privacy & Security → Microphone: confirm Wispr Flow is toggled on. If you changed it, quit and reopen Flow.

3. Confirm input device and volume. Apple menu → System Settings → Sound → Input: select your intended microphone and set the input level so your voice registers clearly without constantly hitting maximum.

4. Rule out Bluetooth or wrong-mic issues. Click Control Center → Sound in the menu bar. If a Bluetooth device such as AirPods is selected but sitting in your pocket or bag, switch to the built-in microphone.

5. Check Flow's microphone picker. Settings → General → Microphone. "Auto-detect" follows your OS default input. Virtual or loopback devices (labeled "(Virtual)") sit under "Show other devices" — selecting one commonly causes silent or empty recordings, and collapsing that section reverts to Auto-detect. Reorder preference with "Edit Ranking", clear disconnected mics with "Forget device", and open the Microphone Quiz via "Find the right mic for you →".

6. Listen to the recorded audio. Click History, find an entry with incorrect transcription, click the three-dot menu, and click Extract audio to save a .wav file. Open it in QuickTime: clipped beginnings or endings, muffled speech, or background noise mean the issue is audio quality, not transcription.

7. Retry the transcription. In the same History entry, click Retry and compare transcripts. A significantly better retry means the original issue was temporary. Retries of recordings longer than 2 minutes show an estimated wait time.

8. Set your preferred languages. Flow icon in the menu bar → Settings → General → Languages → Change. Select the languages you dictate in, or enable Auto-detect for automatic detection across 100+ languages. Flow notifies you when it detects a non-English language or an English regional variant; Canadian-locale users dictating in generic or British English may see a nudge to switch to Canadian English.

Windows

1. Close and restart Flow. Close the Flow window. Right-click the Start button, open Task Manager, and if Wispr Flow is still running, select it and click End task. Reopen the app.

2. Check microphone permissions. Start → Settings → Privacy & security → Microphone: confirm microphone access is on for apps, then scroll down and confirm Wispr Flow has access.

3. Confirm input device and volume. Start → Settings → System → Sound. Under Input, select the correct microphone and set input so your voice registers clearly without constantly peaking.

4. Rule out Bluetooth or webcam mic issues. Start → Settings → Bluetooth & devices to confirm which headset is connected, then in Settings → System → Sound switch from the Bluetooth headset or webcam mic to the built-in laptop microphone and dictate.

5. Check Flow's microphone picker. Settings → General → Microphone. "Auto-detect" follows your OS default input. Virtual or loopback devices (labeled "(Virtual)") sit under "Show other devices" — selecting one commonly causes silent or empty recordings, and collapsing that section reverts to Auto-detect. Reorder preference with "Edit Ranking", clear disconnected mics with "Forget device", and open the Microphone Quiz via "Find the right mic for you →".

6. Listen to the recorded audio. Click History, find an entry with incorrect transcription, click the three-dot menu, and click Extract audio to save a .wav file. Open it in Media Player: cut-off words, low volume, or background noise mean the issue is audio quality, not transcription.

7. Retry the transcription. In the same History entry, click Retry and compare transcripts. Retries of recordings longer than 2 minutes show an estimated wait time.

8. Set your preferred languages. Open Wispr Flow → settings icon → General → Languages → Change. Select the languages you dictate in, or enable Auto-detect for automatic detection across 100+ languages. Flow notifies you when it detects a non-English language or an English regional variant; Canadian-locale users dictating in generic or British English may see a nudge to switch to Canadian English.

iOS

1. Force quit and reopen Flow. On iPhones with Face ID, swipe up from the bottom and pause in the middle; with a Home button, double-press it. Swipe Wispr Flow up to close, then reopen it from the Home screen.

2. Check microphone permissions. Settings → Privacy & Security → Microphone: confirm Wispr Flow is toggled on. If you changed it, quit and reopen Flow.

3. Check your microphone. If you use AirPods or another Bluetooth headset, open Control Center and check the audio input. Disconnect Bluetooth devices and dictate using the built-in iPhone microphone, without covering the top or bottom of the phone.

4. Test in a quieter space. Move away from loud fans, traffic, or other speakers, then dictate a short sentence slightly louder and more clearly than usual.

5. Set your preferred languages. Open Wispr Flow, tap the settings icon, go to language preferences, and select the languages you dictate in.

Android

1. Check your microphone input. Confirm your intended microphone — for example, a hearing aid or Bluetooth headset — is connected and active before recording. If accuracy drops partway through, your input may have switched mid-dictation (commonly a hearing aid handing off to the phone mic): stop the recording, confirm the correct input, and start again.

2. Skip language setup. On Android, Flow auto-detects your dictation language — there is no language selection setting.

3. Report a transcription issue. Find a transcribed entry on the home screen, tap Report, describe what you expected instead, and tap Send (disabled until you enter feedback). Failed transcriptions also show Report, plus Retry Transcript and Delete in the overflow (⋮) menu — Retry appears only when audio was captured, and deleting asks for confirmation and cannot be undone.

Why was my dictation cut off or submitted before I finished?

On desktop, each dictation session has a 20-minute limit, with a warning notification at 19 minutes. At 20 minutes, Flow ends the session and shows a Recover text button — click it to insert the transcript so far into your active text box. For longer content, start a new session.

My dictation stopped on its own, or nothing pasted on iPhone

iOS dictations have a 5-minute maximum; Flow stops recording automatically when reached. If the Flow Keyboard isn't active when the transcript is ready — or if transcription failed — Flow notifies you and saves the transcript in History.

Why does the end of a long dictation look less polished than the beginning?

For dictations of roughly 2 minutes or more, Flow formats the transcript in sections. If a section near the end can't be formatted, Flow pastes everything it formatted and appends the remaining speech as plain text. The content is accurate. If this happens regularly, break long thoughts into shorter sessions.

Why do Bluetooth devices like AirPods cause accuracy issues?

Bluetooth devices often use a lower-quality microphone mode and can clip the beginning or end of your speech. Use the built-in microphone or a dedicated USB microphone instead. On Mac, Flow briefly waits for a Bluetooth headset's audio to stabilize, which can look like a delay at dictation start and stop.

Why does Flow pick the wrong microphone?

Computers can quietly switch input devices when you connect or disconnect hardware; a Bluetooth device in your pocket can become the active mic, leaving Flow with almost no usable audio. Confirm the input device in your system sound settings.

Device names in Flow may not match your system sound settings, because Flow shortens labels: "External microphone (Built-in)" appears as "Headphones (wired)", USB identifiers become "(USB)", and the word "microphone" is dropped. On macOS, Flow can also detect when another app such as Zoom, Teams, Slack, or FaceTime is holding the microphone.

Why does transcription feel worse even though I changed nothing?

The cause is almost always audio-related: a different microphone selected, a new Bluetooth connection, lower input volume, more background noise, or small changes in how you're speaking. Extract the audio from History and listen for quality issues.

Does input volume affect accuracy?

Yes. Low input volume can sound fine to you but be too unclear to transcribe accurately. Flow never changes your microphone input level and only links out to the OS Sound → Input panel — adjust it there so your voice registers clearly without constantly peaking.

Can a large dictionary cause issues?

Yes. Only up to 200 dictionary words (personal plus team, deduplicated) are used per dictation, so very large dictionaries dilute the terms that reach transcription and can cause odd substitutions. Trim your Dictionary to the most important terms and add words back as needed.

Does network quality affect transcription?

Yes. On cellular or unstable connections, Flow compresses audio more heavily, which can reduce accuracy. Brief connection drops are retried automatically; server-side errors are not, so retry from History on a more stable connection.

What happens if transcription fails mid-dictation?

On desktop and iOS, Flow retries the same audio automatically; you may notice a short delay rather than an error, and partial text already shown is preserved. On desktop, that automatic retry is skipped for recordings of 30 seconds or longer — use Retry from History instead. On Android, a failed transcription shows "Failed to transcribe. Saved in the app." with a retry action on the Flow Bubble, which stays tappable offline.

Why is my History section disabled or empty?

A "History disabled" message means your data storage policy is set to Zero Data Retention, either by you in Settings → Data & Privacy or by your organization. Under a delete-after-24-hours policy, entries older than 24 hours are deleted automatically. Zero Data Retention is desktop-only; on iOS, use Auto Delete Transcripts in Settings → Data & Privacy.