VoiceInk repeats text or invents words: the five fixes its own docs give

By · Published

When VoiceInk types a phrase three times, or adds words you never said, the speech model is guessing. VoiceInk's docs call these hallucinations and say they happen when a recording is too quiet, too noisy, too short, or when a model fills a silence. They give five things to try, in roughly this order.

I built YapToText, a dictation app for the Mac, so read what I say about VoiceInk with that in mind. Every fact about VoiceInk here comes from its own docs, read on October 11, 2026.

Download YapToText on the Mac App StoreThis app is completely free. You don't have to make an account. There are no subscriptions, no difficult downloads, and nothing is locked. Made for Apple silicon Macs. Source on GitHub.

1. Turn on Voice Activity Detection

  1. Open AI Models.
  2. Click the gear button.
  3. Select Transcription.
  4. Enable Voice Activity Detection.

It helps local transcription tell speech from silence. The docs note it does not configure every cloud provider's speech detection.

2. Try a different transcription model

Open the Mode you are using and change its transcription model. The docs suggest:

  • Parakeet V3. For fast local dictation on supported Macs.
  • Parakeet V2. If V3 does not suit your audio.
  • Whisper Base or Whisper Tiny. If a larger Whisper model is over-generating.
  • A cloud provider. If local models struggle with your microphone or accent.

The best VoiceInk model page compares them.

3. Check Remove Filler Words

In AI Models, the gear, Transcription, Remove Filler Words strips words such as uh and um, which tidies starts and pauses. The docs add a caution: do not add common words you might really dictate, because they are removed whenever they stand alone.

4. Retry the same recording

After changing the Mode or model, use Retry Last Transcription, set in Settings, Shortcuts. It runs the same saved audio again with your current settings, makes a new History entry, and copies the result without pasting. It needs the audio file to still exist. The shortcuts page lists it with the others.

5. Improve the audio

Poor input makes hallucinations worse. Pick the best microphone in Audio and check its level in System Settings, Sound, Input. The docs say built-in Mac microphones and dedicated external ones usually do better than wireless headset microphones. The VoiceInk microphone page has the settings.

Why speech models do this, and what YapToText does about it

Whisper-family models were trained on captioned audio, so in silence they sometimes produce the kind of text that fills captions. The page on Whisper hallucinations explains it. VoiceInk says it already removes some common artifacts. YapToText does the same kind of filtering on its side: it removes hallucinated speaker labels, stage directions and the phrases silent clips tend to produce, and it throws away a cleanup result that reads like a reply or grows well past what you said. Neither app can promise none will get through.

Download YapToText on the Mac App StoreThis app is completely free. You don't have to make an account. There are no subscriptions, no difficult downloads, and nothing is locked. Made for Apple silicon Macs. Source on GitHub.

The short version

  • First. Voice Activity Detection on.
  • Then. A smaller Whisper model, or Parakeet.
  • Check. Remove Filler Words is not eating real words.
  • Test. Retry Last Transcription on the same audio.
  • Always. A better microphone signal.

If VoiceInk changes these settings, open an issue on GitHub and I will update this page. Every one gets read.

Download YapToText on the Mac App StoreThis app is completely free. You don't have to make an account. There are no subscriptions, no difficult downloads, and nothing is locked. Made for Apple silicon Macs. Source on GitHub.

About the author. Ryleigh Newman is a developer whose hands make typing difficult and who dictates every day with YapToText, the app this site describes. Facts about other apps come from their own sites, with the date they were read. More about the project.

Sources

Questions

Why does VoiceInk repeat text?

The speech model is guessing, usually because the recording is quiet, noisy, short or has silence. Turn on Voice Activity Detection and try another model.

Where is Voice Activity Detection in VoiceInk?

AI Models, the gear button, Transcription, Voice Activity Detection.

Which VoiceInk model hallucinates least?

Its docs suggest trying Parakeet V3 or V2, or a smaller Whisper model such as Base or Tiny if a larger one over-generates.

Can I rerun a bad VoiceInk transcription?

Yes. Retry Last Transcription reruns the saved audio with your current settings and copies the result.

Does a better microphone reduce hallucinations?

Yes. VoiceInk's docs say poor input audio makes them worse.