Skip to main content

Translation and multilingual conversations

Speechlytics can translate recorded transcript segments and live speech. Translation is enabled per account and is separate from transcription-language detection.

Recorded calls​

When translation is enabled for the account, calls in a language other than English get an English translation, produced by a second recognition pass over the audio. Translated segments are returned in the call's translation result by GET /api/v1/calls/{id}/status, with the same timing fields as the original segments, so each translation can be lined up with the passage it translates. Original segments remain available, so you can keep both the source and translated text.

The source language comes from the request or automatic detection. Analytics, keywords and custom questions always use the original-language transcript.

Use GET /api/v1/calls/{id}/status to read translations; list endpoints do not include them.

Live calls​

Live translation can target any supported language. Set translationLanguage when starting a live session to request a fixed target:

{
  "language": 1,
  "translationLanguage": 4,
  "multilingualDetection": false,
  "autoTranslationDirection": false
}

With Live Language Detection enabled, two additional modes are available:

  • multilingualDetection: true re-detects the spoken language on every segment;
  • autoTranslationDirection: true translates each speaker toward the language detected for the other speaker, falling back to the configured target until both languages are known.

These request fields are ignored when the corresponding account features are disabled.

Sensitive content​

When sensitive-content protection and translation are both enabled, Speechlytics protects the translated output as well as the source transcript. See Protect sensitive content.

Practical guidance​

  • Specify a known source language when conversations are consistently monolingual.
  • Use multilingual detection only for calls that genuinely switch languages; repeated detection adds variability.
  • Export both the transcript and the translation when reviewers need to validate ambiguous translations.
  • Transcript search searches the original-language text; with semantic search, a query in one language can still find calls held in another.