Translation

Translate transcripts on your computer

Pick a target language and EdgeSpeak translates each segment in place, next to the source text.

Direct answer

EdgeSpeak translates transcripts and alignment results with an on-device model that covers 38 languages and runs offline.

Translated in place

Each segment keeps its timing. Switch between source, side by side, and translation views.

Tell it what the audio is about

Add background info, translation preferences, dictionary translations, and a translation style.

Export with the translation

Choose how source and translation are arranged when exporting. JSON exports carry translations next to the source text.

Bilingual subtitles

Pick the bilingual SRT preset when exporting: source and translation sit on two lines under the same timestamps, and you can swap their order. In karaoke subtitles only the source lights up word by word; the translation shows as a whole line.

Translate plain text through the local API

Add a translation object to a Responses or Chat Completions request and the on-device translation model handles it. No audio needed, so existing scripts and documents work too. Fields and rules are in the language API docs.

curl http://127.0.0.1:1117/v1/responses -H "Authorization: Bearer $EDGESPEAK_API_KEY" -H "Content-Type: application/json" -d '{"model":"EdgeSpeak/Lattice-2-Flash","input":"Welcome to today’s launch.","translation":{"target_language":"ja","style":"natural"}}'

FAQ

Does translation upload my transcript?

Not with the on-device model, which runs on your computer. If you pick a commercial model service in Providers instead, text is sent to that service.

How many languages?

38, including Traditional Chinese as its own target.

Do I need anything extra?

A valid license or trial, and the translation model downloaded in Models.

Does it need an internet connection?

Not once the translation model is downloaded and ready: on-device translation then runs offline.

Evidence links