Transcription suddenly got worse or feels less accurate

If dictation suddenly loses words, produces garbled text, or sounds less accurate, compare a short test recording with its transcript. This helps separate an audio problem from language, formatting, or transcription errors.

Start with a short comparison

1. Quit and reopen Flow, then dictate a short sentence. If the problem is gone, stop here.
2. Check the microphone. A newly connected headset, webcam, or Bluetooth device in your pocket may have become the active input.
3. Record in another app. Use Voice Memos on Mac/iPhone or Sound Recorder on Windows. Listen for low volume, distortion, or missing first and last words.

If the recording is unclear outside Flow too, fix the microphone or system input first. If audio is clear but Flow's text is wrong, continue to language and transcript checks below.

Check desktop audio

Check	Mac	Windows
Microphone permission	System Settings → Privacy & Security → Microphone	Settings → Privacy & security → Microphone; allow desktop app access
Input device and level	System Settings → Sound → Input	Settings → System → Sound → Input

1. Select your intended microphone and speak. Set the input level so the meter moves clearly without constantly reaching maximum. Flow does not adjust the system input level for you.
2. In Flow, open Settings → General → Microphone and select the intended mic explicitly. Use the built-in or a wired mic to compare against Bluetooth.
3. Avoid an unintended virtual or loopback input. These can route silence or another audio source rather than your voice.
4. Test the same phrase again after each change. Reopen Flow after changing microphone permission.

Bluetooth microphones can delay audio capture and clip opening words. On Mac, allow the mic to become ready before speaking. With a MacBook lid closed, select an external microphone; the built-in mic is unavailable.

For permission errors, disconnected inputs, or silence, see Troubleshooting mic issues. For clipped beginnings, see Missing first words.

Listen to the original and retry

On desktop, open the affected entry in History. If available, use its ⋯ → Extract audio action to save a WAV file and listen in QuickTime or Media Player.

• Audio contains the same gaps: check mic position, Bluetooth, input volume, and background noise.
• Audio is complete and clear: use the entry's Retry action, then compare the result.
• Text exists but did not reach the app: follow text-pasting troubleshooting instead.

Desktop audio recovery requires retained audio and an eligible recording. Retry is unavailable during active dictation; recordings at least five seconds long and under 14 days old can be eligible. Longer retries can show an estimated wait. Retention policies may remove audio sooner or disable history. See Retry failed transcriptions for platform-specific recovery.

Check phone audio

iPhone

Reopen Flow and confirm microphone permission in iOS Settings. Disconnect Bluetooth for a comparison with the built-in mic, keep the phone's microphones uncovered, and test in a quieter space. Flow's built-in-mic preference is covered in microphone troubleshooting.

A capture failure needs a fresh dictation; it cannot recover audio that was never recorded. Audio download is unavailable on iOS. Use History's available Retry or Report actions for a saved recording.

Android

Confirm that your intended microphone or hearing aid is connected before recording. If accuracy drops partway through, an input handoff may have occurred: stop, confirm the active input, and start again.

For an incorrect transcript, tap Report on its home-screen entry, explain the expected text, and choose Send. A failed entry can offer ⋮ → Retry Transcript when audio exists. Deleting an entry requires confirmation and cannot be undone.

Check language and repeated substitutions

Select the languages you actually use. On desktop, use Settings → General → Dictation Languages; on iOS, Settings → Set language; on Android, Settings → Languages. Follow language setup for saving choices and regional variants.

If names or technical terms are repeatedly wrong, review your dictionary words and corrections. Dictation language settings do not control Notetaker meetings.

If words are present but the end of a long dictation is less polished, formatting may not have completed. Retry when available, or divide long thoughts into shorter recordings. See accuracy and known limitations for formatting and session limits.

Report a persistent change

Contact support when clear, complete audio repeatedly produces missing or incorrect text after these checks. Use Help → Talk to support on desktop or an entry's Report action. Include your platform, microphone, selected languages, a specific recording, what you said versus the result, and the checks you completed.