Every voice in the service, told apart.

Few services are one voice from start to finish. A pastor preaches, an elder prays, someone reads the Scripture, a host gives the announcements, and in a bilingual church an interpreter may repeat every sentence. Captions that run all of them together are hard to follow, above all when two voices say the same thing in two languages.

Speaker detection (engineers call it diarization) hears where one voice ends and the next begins, so each person’s words stay together on the page.

Once a second voice has spoken, every sentence on the listener’s page carries a thin line in its speaker’s colour along its edge, joined across one speaker’s turn. Where the voice changes, a small label names the speaker by number in the listener’s own language: Speaker 2, Orador 2, 발언자 2, 講員 2. When one person speaks throughout, as in most sermons, there is no rail at all.

The labels are numbers, never names: Voclyn hears that the voice changed, not who is speaking.

Speaker detection works together with language switching. A pastor who moves from Korean to English keeps the same colour, and each sentence is still translated from the language it was spoken in.

In our test on 29 October 2026, one computer voice read a sentence in Mandarin and the next in Korean, and Voclyn kept it as one speaker. Where the voice changes, Voclyn closes the sentence, so the next speaker starts a new caption.

The console at the sound desk shows each voice in its own colour as it hears it. The operator can tap a voice to call it Pastor, Reader, Interpreter or Host; those names stay at the desk, and phones keep the numbers.

After the service, the saved transcript marks every change of speaker too, in the dashboard and in the downloaded text file.

When an interpreter repeats each sentence on stage, Voclyn hears two voices, and the interpreter gets their own colour. Today the interpreter is captioned like everyone else: their words are translated for every phone, so readers see each sentence twice, once from the pastor and once from the interpreter.

Interpreter mode, for routing an interpreter’s words, is coming next. Until then a church can let the phones carry every language, so the sermon is said once, or send Voclyn a mix from the mixer that leaves the interpreter’s microphone out.

• Rooms that include Cantonese. They use their own speech engine, and Voclyn doesn’t tell voices apart there yet. Captions and translation work as usual.
• Similar voices. Speaker detection works from the sound alone, so two similar voices can be read as one. In our test, two different women’s computer voices, one reading English and one Spanish, came out as one speaker.
• Names. Voclyn doesn’t know who anyone is, and service audio is never stored.
• The big screen. Speaker colours are on phones and at the desk; the projector shows whole sentences without them.
• More than four voices. The colours repeat after the fourth; the number in the label keeps counting.