EdgeSpeak Translate

Translate a timed transcript into another language. This skill is agent-driven: you write the translations. There is no edgespeak-cli translate, no translation API, no external MT service — the quality comes from your own language ability plus the rules below. That also keeps the whole pipeline local: the audio was transcribed on-device, and the text should not be shipped to a cloud translator either.

It fits between the other skills: edgespeak-transcribe / edgespeak-align produce the timed source, this skill adds the target language, and edgespeak-broadcast can speak the result as a dub.

Inputs to confirm

• Source transcript — an EdgeSpeak JSON (segments[] with start / end / text, optionally words[]) or an .srt. A plain wall of text has no timing; run edgespeak-segment on it first if per-sentence output is wanted.
• Target language.
• Output shape — translation-only SRT, bilingual SRT (source line + target line), or JSON with a translation field per segment (best when anything downstream needs the timings).
• Will it be spoken? If the translation feeds edgespeak-broadcast for dubbing, the length budget below is a hard constraint, not a preference. Ask this before translating — it changes every line.

Invariants

These hold for every segment. A violation means the output no longer matches the audio.