Repetitions and hallucinations in transcription

Sometimes QuickWhisper may produce repetitive segments or text that is not present in the original media file. This is known issue with Whisper models. This typically happens with:

• Smaller size Whisper models (Tiny, Small, Base)

• Recordings containing prolonged silence

• Recordings has background noise

• Recordings has overlapping speakers or unclear speech

Recommended Solutions

Step 1: Use the optimal model for your language

Try more advanced Large V3 models for your language if available through the model manager. These newer models generally provide better accuracy and less prone to repetition and hallucination.

Step 2: Enable audio pre-processing

Audio pre-processing performs various optimizations including removing silence and boosting voice levels in the audio, which can significantly improve overall transcription quality.
Go to Settings → Advanced → Optimize audio and ensure this feature is enabled.

Step 3: Use manual editing features

Finally, if the problem persists, try powerful editing capabilities that let you:

• Delete repeating segments entirely

• Trim segments before/after the current segment

• Split segments that weren't properly divided