3: Subtitles and dubbing

The audio track is not just the clock; it is text waiting to happen. Presentation transcribes it, translates it, and can re-voice it.

3.1: Transcription

Whisper transcribes the audio into timed segments, with the source language detected automatically. The result is a subtitle track locked to the audio clock and an SRT file you can download. Because the segments are timed, they double as the spine for everything else: both dubbing and the AI director place their work against them.

3.2: Translation

From the transcript, a GPT call translates the subtitles into another language while preserving the timings. Each language becomes its own SRT and its own track, and the player gets a language switcher so a viewer picks their subtitles at play time.

3.3: AI dubbing

Subtitles are read; dubbing is heard. Presentation groups the transcript by sentence and generates TTS narration in a preset voice, placed back on the transcript timeline without cutting mid-sentence. The dub is attached as a language alternative to the audio, so the same presentation can play in the original voice or a dubbed one.

3.4: Language alternatives everywhere

Language is not only the audio. Every media item can carry per-type language alternatives: a different image, a different video or a different caption per language. The player's language switch then swaps the audio, the subtitles and the per-item alternatives together, so one presentation is genuinely multilingual rather than one presentation per language.