Quality
Tested, dated and published.
On 29 October 2026 we streamed sentences in English, Spanish, Mandarin and Korean through Voclyn’s live system and timed every caption. The words appeared on the phones as they were spoken, and each sentence was translated right after it ended, in every language we tested.
This page gives the method, the limits and the raw data, so anyone can check what we say.
What we found
- The original words appeared on the phones as they were spoken.
- Each sentence was translated right after it ended, in English, Spanish, Mandarin and Korean.
- Speech recognition made 1 error in 115 English words (“two” written “2”).
- Every part was labelled with the right language, including two sentences that switched language midway.
How we measured
- We wrote short sermon-like sentences and had a computer voice read them: English, Spanish, Mandarin and Korean, and two sentences that switch language in the middle.
- Each recording was trimmed to the voice, with 0.6 seconds between sentences, as when a preacher barely pauses.
- A script streamed the audio into the live system at speaking pace, through the same public demo anyone can try on our site, sending each tenth of a second when a microphone would have finished capturing it.
- Every caption and translation was timed against the audio: from the moment a sentence ended to the moment its translation arrived.
- Every timing is public, sentence by sentence: raw timings (CSV).
What this doesn't show
This is one test with a clean computer voice, two runs and one internet connection. A pulpit microphone in a room with an echo, music in the background and a preacher who never pauses will be slower; and so is a preacher who barely pauses.
We haven’t measured Cantonese rooms (they use a separate speech engine), the AI voice, or busy church Wi-Fi with a hundred phones. Next we will measure real services at pilot churches, with their permission, and publish those numbers here with their dates.
This is not a comparison with other products. We haven’t tested them, so we don’t claim to be faster than anyone.
Accuracy
Every translation on a listener’s phone is labelled AI, except Bible readings shown from a published Bible, which are labelled “Published text · not AI”. A passage is shown only when it was named and what follows is mostly its words; a paraphrase gets no card, because a wrong verse is worse than none.
We don’t publish an accuracy percentage yet. The next test checks what churches care about most: pastors’ names, church words and Bible books, with and without the glossary.
Questions
What churches ask.
Something else on your mind? Write to us at hello@voclyn.com.
We don't claim it: we haven't tested other products. We publish how we test our own, with the method and raw data.
A sentence can't be translated well until it's finished, because the verb or the meaning may come last. The original words appear while they are spoken; the translation follows each sentence.
Yes. The demo on our site runs the same system: press start and speak, in one language or several.
Try it this Sunday.
Set up your church in a minute and hear it live with your own voice. Your QR code keeps working after the free month.
Your first month is on us: the Church plan, normally $99 a month, with unlimited listeners, translation hours and AI voiceover. No card; then choose a plan, or stay on Free.