`<Transcription>`

Live transcription of the call, delivered to your webhook while the document keeps executing.

Used inside<Start>, <Stop>.

Example

<Response>
  <Start>
    <Transcription statusCallbackUrl="https://app.example.com/transcripts"
                   partialResults="true" track="both_tracks"/>
  </Start>
  <Dial>+18005551234</Dial>
</Response>

Attributes

Attribute	Default	Behaviour
`name`	`unset`	Operator-friendly identifier for the session, echoed on every callback as Name.
`languageCode`	`en-US`	BCP-47 language tag passed to the transcription engine. The platform's built-in engine transcribes English only — a non-English tag with no cloud engine configured is refused with a transcription-error; configure deepgram, google, aws, or azure for other languages.
`transcriptionEngine`	`account default`	Selects the transcription engine (case-insensitive): deepgram, google, aws, or azure. An unrecognized value — including auto — refuses the session with a transcription-error naming the engine; when unset, the account default (or the platform's built-in engine) is used.
`speechModel`	`engine default`	Recognizer model hint, mapped per engine (for example the short/long aliases select the engine's latest short-form or long-form model).
`partialResults`	`false`	true also delivers interim transcripts to the status callback; the default delivers final transcripts only.
`track`	`both_tracks`	Which audio to transcribe: inbound_track, outbound_track, or both_tracks. Passed to the engine as a hint; engine support varies.
`statusCallbackUrl`	`unset`	URL receiving every lifecycle event — transcription-started, transcription-content, transcription-stopped, transcription-error — as form-encoded requests. Spelled exactly statusCallbackUrl: statusCallback is not accepted on <Transcription>.
`statusCallbackMethod`	`POST`	HTTP method for status callbacks.
`hints`	`unset`	Comma-separated phrases passed through to the engine to bias recognition.
`profanityFilter`	`true`	Mask profanities in the transcripts (passed through to the engine).
`enableAutomaticPunctuation`	`true`	Let the engine punctuate transcripts (supported by the deepgram and google engines).
`enableProviderData`	`false`	true attaches the engine's raw provider payload as TranscriptionProviderData on content events (deepgram engine only).
`inboundTrackLabel`	`unset`	Label echoed for the inbound track on the transcription-started event.
`outboundTrackLabel`	`unset`	Label echoed for the outbound track on the transcription-started event.
`intelligenceService`	`unset`	Accepted for compatibility; no such product exists on the platform, so setting it logs an unsupported-feature notification and has no effect.
`conversationConfiguration`	`unset`	Accepted for compatibility and silently ignored.
`conversationId`	`unset`	Accepted for compatibility and silently ignored.

Notes

Attribute names are checked exactly, including case. An attribute outside the accepted set — statusCallback and statusCallbackEvent included — refuses the whole transcription with a transcription-error naming the parameter.
Callback fields on every event: AccountSid, CallSid, TranscriptionSid, Timestamp, SequenceId, TranscriptionEvent. transcription-started adds the session settings (TranscriptionEngine, Track, LanguageCode, PartialResults, and Name when set). transcription-content adds LanguageCode, Track, TranscriptionData (JSON with the transcript, plus its confidence on final results), Final (true or false), and Stability on interim results. transcription-error adds TranscriptionErrorCode and TranscriptionError.
Unlike speech-mode <Gather> (listen once, then return), <Start><Transcription> runs long-lived and in parallel with the rest of the document.
One active transcription per call: a second <Start><Transcription> replaces the first. <Stop><Transcription/> ends it early; otherwise it ends at hangup.