All docs

AI features

AI scripture detection

How Neopresenter listens to your preacher and projects the right verse in under 600 milliseconds.

Scripture detection is Neopresenter's signature feature. Here's a technical walkthrough of how it works and how to tune it for your context.

The pipeline

  1. Audio capture — A 24kHz audio stream is captured from your selected microphone (separate from the broadcast audio path, so monitoring effects don't interfere).
  2. Transcription — A fine-tuned Whisper model produces a rolling transcript with word-level timestamps.
  3. Reference detection — A second model (we call it neo-bible-1) scans the transcript for direct references, paraphrased quotes, and topical allusions across 31 Bible translations.
  4. Confidence scoring — Each candidate verse is scored from 0–1. By default, anything above 0.72 is auto-projected.
  5. Rendering — The matched verse is rendered using your active theme and pushed to projection within 600ms of being detected.

Tuning the confidence threshold

Open Settings → AI → Detection sensitivity. The slider exposes the confidence threshold:

  • High (0.85) — Only display verses when the preacher quotes them directly. Best for liturgical services.
  • Medium (0.72, default) — Display direct quotes and confident paraphrases. Best for expository preaching.
  • Low (0.55) — Display anything with a plausible scriptural connection. Great for topical sermons, but expect occasional false positives.

Handling paraphrases

neo-bible-1 is trained on a dataset of 400,000 paraphrased scripture references from real sermon transcripts. It will surface the underlying verse even when the preacher says something like "Paul tells us that nothing can separate us from God's love" (Romans 8:38–39) without quoting it directly.

Translation preferences

Configure your church's preferred translation in Settings → AI → Translation. When multiple translations would match, the AI defaults to your preferred one. You can also configure per-service overrides for guest speakers.

Privacy

Audio is processed in your edge region and is never stored unless you explicitly enable transcription archival. Transcripts, when stored, are encrypted at rest and only accessible to members of your workspace.

Was this helpful?