No AI Is Perfect

Let’s get this out of the way: Vies will occasionally suggest the wrong verse. It will also miss verses that were clearly referenced. This is the reality of AI-powered detection in a live environment, and pretending otherwise would waste your time.

What matters is how often it gets things right, how it handles the cases where it doesn’t, and what controls you have to keep bad matches off your screens. This article covers all three.

When It Works Well

Vies performs best in the conditions that describe most Sunday morning sermons:

  • Direct quotes. When the minister says “for God so loved the world that He gave His only begotten Son,” Vies recognizes John 3:16 almost immediately. The closer the words match the actual Bible text, the higher the confidence.
  • Clear spoken references. “Turn to John chapter 3, verse 16” or “look at Romans 8:28 with me.” Vies has a dedicated parser for these patterns and handles them reliably.
  • Moderate speaking pace. A conversational preaching speed gives the system time to process each reference before the next one arrives.
  • Clean audio. A direct feed from the mixing desk or a well-positioned microphone eliminates most of the noise that causes transcription errors.

Under these conditions, Vies catches the large majority of verse references without intervention.

When It Struggles

These are the situations where you should expect lower accuracy:

  • Vague allusions. If the minister says “remember the story of the prodigal son” without naming Luke 15 or quoting the text, Vies has very little to work with. It does not try to guess at indirect references.
  • Heavy or unfamiliar accents. The speech recognition handles a wide range of accents, but some regional speech patterns produce transcription errors that cascade into missed or wrong verse matches.
  • Rapid-fire references. A preacher who lists six cross-references in ten seconds will overwhelm the detection pipeline. Some of those references will be missed or arrive late.
  • Background noise and poor mic placement. Loud worship bands bleeding into a room mic, HVAC hum, congregation noise — all of it degrades the transcript, which degrades verse detection. Audio quality is the single biggest factor in accuracy.
  • Paraphrased verses. If the minister loosely restates a verse in their own words without quoting or naming it, the match confidence drops. Heavy paraphrasing often falls below the detection threshold entirely.

None of these are bugs. They are inherent limitations of working with live audio and natural speech. Knowing where the boundaries are helps you set realistic expectations.

The Confidence System

Every verse match that Vies produces comes with a confidence score. This is the system’s estimate of how likely the match is correct.

Low-confidence matches are filtered out before they ever reach your screen. You control where that cutoff sits. In the Scripture Listener settings, you will find a minimum confidence threshold:

  • Raise it if you are seeing false positives — random verses appearing that the minister did not reference. A higher threshold means fewer matches overall, but the ones that get through are more likely correct.
  • Lower it if Vies is missing references that were clearly spoken. You will get more matches, but you will also see more occasional false positives.

The default threshold works well for most preachers. Adjust it based on what you observe in your specific setup.

The Cooldown System

Preachers repeat themselves. A minister might say “Romans 8:28,” read the verse aloud, then say “Romans 8:28 tells us…” all within thirty seconds. Without protection, that would cause the same verse to fire three times.

Vies uses a cooldown timer. Once a verse is detected and displayed, the same verse is suppressed for a configurable number of seconds. This prevents screen flickering during repeated references and keeps the output clean.

There is also a cap on how many verses can fire per transcript update, so a rapid list of cross-references does not cause a cascade of rapid screen changes.

The Operator Is Still There

Vies is a tool, not a replacement for the person at the projection desk.

Detected verses appear in a suggestion list or overlay. The operator can see what Vies is about to display, dismiss a false positive with one click, or manually search for a verse that was missed. The software assists; it does not lock anyone out of control.

This is a deliberate design choice. In a live service, a human with context will always make better judgment calls than an automated system. Vies handles the routine work of listening, searching, and suggesting. The operator handles the exceptions.

Improving Accuracy

If you are seeing more misses or false positives than you would like, here are the most effective adjustments, roughly in order of impact:

  1. Improve your audio source. A direct feed from the mixing desk into a USB audio interface will outperform any room microphone. If a direct feed is not possible, move the mic closer to the speaker and away from the PA system. See the microphone setup guide for specific recommendations.
  2. Adjust the confidence threshold. Start at the default and move it up or down in small increments based on what you observe during a live service.
  3. Reduce background noise. Muting the worship band’s channels on the monitor mix during the sermon, turning off nearby HVAC vents, and asking the congregation to settle all help the transcription quality.
  4. Test before Sunday. Run Vies during a rehearsal or with a recorded sermon. This lets you dial in settings without the pressure of a live service.
  5. Enable Voice Activity Detection. VAD filters silence, hymns, and background noise before they reach the speech engine. This is one of the most effective ways to reduce false positives — turn it on in Scripture Listener settings.
  6. Choose a higher engine tier. The Best speech engine (1.5 GB, GPU required) produces more accurate transcription than the Lite engine. Better transcription means better verse detection.
  7. Try Cloud AI mode. If you have a Gemini API key, Cloud AI adds a second layer of verse identification. No audio leaves your machine — only transcript text is sent.

The common thread is that better audio in means better detection out. Most accuracy problems trace back to the microphone, not the software.

Get Started

If you have not tried Vies yet, the Getting Started guide walks you through installation and your first test run in about ten minutes. And if you want to go deeper on microphone setup, the microphone guide covers everything from budget USB mics to mixing desk integration.

Vies will not be right every time. But with a decent audio source and sensible threshold settings, it handles the repetitive work of verse detection reliably enough that your operator can focus on the exceptions instead of searching for every single reference by hand.