Repeated Text or Hallucinations
Reduce repeated, bracketed, or imaginary text in transcriptions.
What This Looks Like
Hallucinations are words, phrases, tags, or repeated text that were not clearly present in the audio. They can happen when the recording is too quiet, too noisy, too short, or when a model guesses during silence.
VoiceInk already removes some common transcription artifacts, but model choice and audio quality still matter.
Turn On Voice Activity Detection
Voice Activity Detection helps the transcription model ignore silence and low-confidence audio.
- Open AI Models.
- Click the gear button.
- Select Transcription.
- Enable Voice Activity Detection.
Try a Different Transcription Model
Open the Mode you are using and change its transcription model.
Good options to try:
- Parakeet V3 for fast local dictation on supported Macs.
- Parakeet V2 if V3 is not the best fit for your audio.
- Whisper Base or Whisper Tiny if a larger Whisper model is over-generating.
- A cloud transcription provider if local models struggle with your microphone or accent.
See AI Models and Recommended Models.
Check Filler Words
Open AI Models -> gear -> Transcription -> Remove Filler Words.
Removing filler words such as uh and um can clean up starts and pauses. Avoid adding common words you may actually dictate, because they will be removed whenever they appear as standalone words.
See Filler Words.
Use Retry Last Transcription
If a recording produced repeated or imaginary text, change the Mode or model settings, then use Retry Last Transcription from Settings -> Shortcuts. This runs the same audio again with your current configuration, so you can compare models without recording again.
Improve the Audio Source
Poor input audio can make hallucinations worse. Use Audio to select the best microphone, and check the input level in System Settings -> Sound -> Input.
Built-in Mac microphones and dedicated external microphones usually work better than wireless headset microphones.