Understanding Transcription Accuracy
AI transcription is powerful but imperfect. Accuracy depends heavily on audio quality, background noise, speaker clarity, and accents. Here is how to diagnose and improve transcription quality.
Audio Quality Assessment
Before troubleshooting the transcription, assess the source audio:
- Play the video and listen carefully. Can you understand the speech?
- Check the audio waveform in the timeline. Healthy speech appears as regular amplitude patterns. Flat lines indicate silence; solid blocks indicate noise.
- If you cannot clearly understand the speech by ear, the AI will also struggle.
Improving Results with Poor Audio
If the audio is noisy or unclear:
- Enable Enhanced Audio Processing: Go to Settings > Transcription > Enable Enhanced Audio Processing. This applies noise reduction, dynamic range compression, and speech enhancement before transcription.
- Adjust the audio channel: Some body cameras record stereo audio where one channel is clearer. Try selecting a specific channel in Settings > Transcription > Audio Channel.
- Lower the speed: In Settings > Transcription > Processing Speed, choose "Accuracy" mode instead of "Balanced." This uses a larger model and takes longer but produces better results.
Tip: Enhanced Audio Processing adds roughly 30% to the transcription time but can significantly improve accuracy in noisy environments like traffic stops, protests, or indoor environments with echo.
Handling Multiple Speakers
When multiple people are talking simultaneously or in rapid succession:
- Enable Speaker Diarization: Settings > Transcription > Speaker Diarization. This labels different speakers (Speaker 1, Speaker 2, etc.).
- Set expected speaker count: If you know how many people are present, set the count in the diarization settings. This helps the model separate voices more accurately.
- Review overlapping speech: Moments where speakers talk over each other may have reduced accuracy. Flag these sections for manual review.
Accents and Dialects
If the speakers have strong accents or use regional dialects:
- Set the language explicitly rather than using auto-detect.
- Consider the "Accuracy" processing mode for better handling of non-standard pronunciation.
- Manual correction may be needed for specialized terminology or proper nouns.
Re-Running Transcription
If you change settings and want to try again:
- Right-click the video in the Evidence Panel.
- Select Re-Transcribe.
- The previous transcript is archived (not deleted) and a new one is generated.
You can compare multiple transcript versions in the Findings Panel by selecting Show Transcript History.
Warning: No AI transcription is 100% accurate. Critical sections — especially those you plan to cite in court filings — must be manually verified by listening to the actual audio. Mark verified sections using the checkmark tool in the transcript editor.