When live translation falls behind the speaker, our system speeds up the delivery rather than skipping ahead to catch up. That's a deliberate choice: the congregation hears every word of the message, not a summary. It matters because in most systems, falling behind means dropping content — professional interpreters treat 100 to 120 words per minute as the range for keeping quality intact (AIIC), and beyond that, things get cut.
Here's what this post covers, at a glance:
- What "latency" actually means in live sermon translation
- Why most systems skip words when they fall behind
- Why we chose to speed up instead
- What that means for you and your congregation
- How to keep latency low in the first place
What does latency mean in live translation?
Latency is simply the gap between the moment a word leaves the preacher's mouth and the moment it reaches a listener's ear in their own language. A little delay is unavoidable — the system has to hear the phrase, understand it, translate it, and speak it back out. That's true of a human interpreter in a booth too. The question isn't whether there's a delay. The question is what happens when that delay grows.
When a preacher gets fired up and the words start flowing faster than the translation can keep pace, something has to give. And that "something" is where the real design decision lives.
Why do most systems skip words when they fall behind?
Because skipping is the easy way out. When a system — or a tired human interpreter — falls behind fast speech, the simplest fix is to jump ahead and summarize, leaving gaps in the middle.
The research on this is sobering. A large bilingual speech corpus study found that when speech rate climbs above 7 syllables per second, roughly half of the content words go missing in the translation (Association for Computational Linguistics). Even at slower rates, about 20% of translations were missing. The same study confirmed that missing content spikes when speech is fast and the delay is large.
Read that again. Half the meaningful words — gone. Not because anyone chose to remove them, but because catching up meant leaving them behind.
Why did we choose to speed up instead?
This wasn't a whiteboard theory. It came out of a live service where the translation fell behind and we watched, in real time, what "catching up by skipping" does to a message. It gutted it. A sentence about grace turned into a fragment. So we changed the rule.
When playback falls behind, our system compresses the delivery — talks a little faster — instead of jumping ahead. The listener still hears the complete thought. Nothing is quietly dropped to make the clock happy.
It's a theological commitment, not just an engineering one
We kept coming back to one conviction: the preacher chose those words. Every one of them. When you skip, you appoint an algorithm to decide which parts of the sermon your congregation doesn't need to hear. That's not a call software should make. A congregation deserves the whole message, and "every word is spoken" is a promise built into how Jesus Speaks works.
Isn't faster playback also a problem for accuracy?
Fair question. Faster isn't free — the interpreting profession knows this well. That's exactly why 100 to 120 words per minute is treated as the sweet spot for quality, with accuracy dropping as speech gets faster (AIIC).
So we don't speed up to a chipmunk sprint. We nudge the pace just enough to close the gap without shredding intelligibility, then settle back to normal once we've caught up. It's the difference between a driver easing back to the speed limit and one who slams the brakes and skips three exits. The U.S. broadcast captioning rules actually frame this same tension — the FCC weighs accuracy and completeness against timing lag on a case-by-case basis rather than allowing arbitrary omission (47 CFR § 79.1). Completeness matters. We agree.
Who does this actually serve on a Sunday morning?
More people than you might guess. More than 1 in 5 people age 5 and older in the United States — about 22% — spoke a language other than English at home over the 2017–2021 period (U.S. Census Bureau). That's roughly 68 million people, and about 29.6 million of them have limited English proficiency per 2023 Census/ACS data.
For ministries, that's not an abstraction. It's the family in the third row who catches maybe 60% of the sermon in English and fills in the rest with hope. Among people who speak another language at home, about 61% speak Spanish (Census Bureau, June 2025) — and most Hispanic Protestant pastors in the U.S. conduct worship in Spanish (Lifeway Research). When translation drops words for that listener, it drops them right where understanding was already fragile.
How do you keep latency low in the first place?
Speeding up is the safety net, not the strategy. The best latency is the kind you never notice, and a few practical things help:
- Feed the system your vocabulary ahead of time. Upload sermon notes or type in a line of context, and proper names, place names, and ministry terms improve in both transcription and translation. A visiting pastor's name or a town like "Braga" won't get mangled.
- Encourage a natural preaching pace. Not slow — natural. The 100–120 words-per-minute range that keeps human interpreters accurate helps machine translation too.
- Use the right delivery path for your people. Younger folks may read or listen on a phone; older members who won't scan a QR code can be served through a radio transmitter into an existing hearing-assist loop. Same message, no one left out.
Why this is worth caring about
A sermon isn't a status update. It's a shepherd handing bread to a hungry congregation, and you don't get to keep the crust and toss the middle. When we decided to speed up instead of skip, we were really deciding that the person listening in Spanish, Portuguese, or any of the 83 languages we can carry at once deserves the same meal as the person in the front row who speaks English.
That's the whole reason this was built in the pews and not in a lab. Every feature traces back to a real moment in a real service where something broke and someone got left out. This one traces back to a message that deserved to arrive whole.
If you lead a church or ministry and want your whole congregation to hear the whole message, you can subscribe to get started and bring live translation into your services.