Translation
One language pair per stream; direction is resolved per segment.
Give the stream one language pair. Translation streams alongside the transcript rather than waiting for the sentence to finish.
Request
Fields
| Field | Type | Description |
|---|---|---|
languages | string[]? | Language pair, e.g. ["en","zh"]. Omit the block to disable translation. |
languagesis a pair, not a source and a target. A segment recognised aszhis translated toen; everything else —en, or any third language that appears mid-session — is translated tozh. The second entry acts as the fallback, as the default translation target. One-way lectures, two-way conversations and mid-session language switches all use the same configuration.
Two-way translation and other languages
["en", "zh"] means English → Chinese and Chinese → English. Other detected languages, such as Japanese, translate into the second language (Chinese). To allow third languages, set transcribe.strict: false and transcribe.identifyLanguage: true (or use languages: ["auto"]). Fallback translations may arrive after the original transcript; a provider failure can leave a segment without a translation.