Narration

A film's caption is what the narrator says. Give a course a voice, and Kapi reads every caption in it aloud.

Give a course a voiceCopy Anchor Link

In kapi-works.json, add voice and say "sdk": "2.2":

{
  "sdk": "2.2",
  "voice": { "name": "Sulafat" }
}

Kapi's voices are Google's ready-made Gemini voices. The same name always gives the same voice.

A course without voice has no voice: its films show captions only. A course Kapi makes with a film gets Kapi's voice, Sulafat. On the course's page in Studio, under Voice, you can choose another voice or none.

Kapi checks this: voice_sdk.

What Kapi recordsCopy Anchor Link

• Every film caption, in the course's language, with the accent of that language: Southern British English for en-GB.
• Each caption once for your company. Another course with the same words in the same voice plays the same recording.
• A changed word records its caption again. Bold, emphasis and links change nothing a learner hears, so they record nothing.
• Kapi refuses a placeholder such as {company.name} in a caption, as it records the caption once for everyone. Say "your company" instead. Kapi checks this: narration_placeholder.

Write captions to be heard: short sentences, numbers as people say them, and no symbols the narrator would have to read out.

How Kapi checks each recordingCopy Anchor Link

Kapi levels every line to -18 LUFS, with peaks under -1 dBTP, so every line and every voice sounds as loud as the others. Then it listens back: Gemini writes down the words it hears, and they must match the caption. Kapi also checks that the line isn't cut off, has no long silence, and isn't too fast or too slow.

A line that fails records again, up to three takes. A line that fails three times is flagged with the reason, and only its caption shows. On the course's page you can record it again.

The Voice section on the course's page shows each caption and how it stands: waiting, recording, recorded with its length, or the reason Kapi stopped it.

What learners getCopy Anchor Link

• The film plays to the recording. A scene lasts its seconds, or its recording and 0.8 seconds, whichever is longer. Each caption sentence shows while it is said.
• An AI voice switch beside Captions turns the voice off; the captions go on. Kapi keeps the choice with the course.
• The voice starts with the learner's first tap, as browsers ask.
• Publishing doesn't wait for the voice. A caption Kapi hasn't recorded yet shows as a caption only.

CostCopy Anchor Link

Recording uses your company's AI credits, at what Gemini charged for each line: the voice and the listening back together came to about 2 US cents a minute of narration on 19 October 2026. Kapi records each caption once.