Transcription

The <Transcription> verb enables real-time speech-to-text transcription for the call.

Attributes

ATTRIBUTE	DESCRIPTION	OPTIONS	DEFAULT
`language`	Language to use for speech recognition. See languages in transcription_engine_config.		`en`
`interimResults`	Whether to send also interim results. If set to false, only final results will be sent. Applies to transcriptionEngine A only.		`false`
`transcriptionEngine`	Engine to use for speech recognition. Available engines: Google (default), Telnyx, Deepgram, Azure, xAI, AssemblyAI, Soniox, Speechmatics, Parakeet, Humain, Reson8, Cohere, or a community engine (use its registered name, and send `apiKeyRef`). Each engine supports different models and languages - see model attribute and examples below. Legacy values “A” (maps to Google) and “B” (maps to Telnyx) are supported for backward compatibility.		`Google`
`transcriptionTracks`	Indicates which leg of the call will be transcribed. Use inbound for the leg that requested the transcription, outbound for the other leg, and both for both legs of the call.	`inbound`, `outbound`, `both`	`inbound`
`transcriptionCallback`	URL that tells Telnyx where to make its GET or POST requests with transcription data.		-
`transcriptionCallbackMethod`	HTTP request type used for transcriptionCallback.	`GET`, `POST`	`POST`
`model`	Optional model to use with the specified transcription engine. Format is `vendor/model-name` — e.g. `deepgram/nova-2`, `deepgram/nova-3`, `azure/fast`, `assemblyai/universal-streaming`, `soniox/stt-rt-v4`, `speechmatics/standard`, `speechmatics/linden-1` (preview, explicit language required), `nvidia/parakeet-v3`, `omi-health/omi-med-stt-v1`, `xai/grok-stt`, `humain/realtime`, `reson8/turns`, `cohere/ar-stt`. Deepgram also offers `deepgram/nova-2-medical`, `deepgram/nova-2-phonecall` and `deepgram/nova-3-medical` (English only), and `deepgram/flux` / `deepgram/flux-multi`. The vendor must match `transcriptionEngine`. If not specified, the engine’s default model will be used. On Deepgram, defaults to `deepgram/nova-3`. For a community engine, this is the name of one of the engine’s models (1-32 characters: lowercase letters, digits, hyphens, underscores and dots); omit it to use the engine’s default model. Not a vendor/model pair. See transcription models in transcription_engine_config.		-
`hints`	Hints to improve transcription accuracy. On Deepgram, this maps to the Nova-2 keyword biasing feature and is supported only on Nova-2 models (`deepgram/nova-2`, `deepgram/nova-2-medical`, `deepgram/nova-2-phonecall`); it is silently dropped on Nova-3 and Flux models (use `keyterms` instead). Accepts a comma-separated string.		-
`keyterms`	Deepgram keyterm prompting. Biases recognition toward domain-specific terms or brand names. Supported only on Nova-3 and Flux models (`deepgram/nova-3`, `deepgram/nova-3-medical`, `deepgram/flux`, `deepgram/flux-multi`); silently dropped on Nova-2 models (use `hints` instead). Accepts a comma-separated string.		-
`smartFormat`	Disable Deepgram’s smart formatting so the transcript stays lowercase with no punctuation. Deepgram-only; silently dropped on other engines.		`true`
`apiKeyRef`	Reference to the API key for authentication. See integration secrets documentation for details. The parameter is optional as defaults are available for some regions. Required with a community engine: the Integration Secret holding the key the engine expects.		-
`region`	Region to use with the specified transcription engine. Required for Azure. See regions in transcription_engine_config.		-

Examples

Example 1: Basic transcription using Google (default)

<?xml version="1.0" encoding="UTF-8"?>
<Response>
  <Start>
    <Transcription language="en" interimResults="true" transcriptionCallback="/transcription" />
  </Start>
</Response>

Example 2: Google with phone call optimized model

<?xml version="1.0" encoding="UTF-8"?>
<Response>
  <Start>
    <Transcription
      transcriptionEngine="Google"
      model="phone_call"
      language="en-US"
      interimResults="true"
      transcriptionCallback="/transcription" />
  </Start>
</Response>

Example 3: Telnyx engine with Whisper model

<?xml version="1.0" encoding="UTF-8"?>
<Response>
  <Start>
    <Transcription
      transcriptionEngine="Telnyx"
      model="openai/whisper-large-v3-turbo"
      language="es"
      transcriptionCallback="/transcription" />
  </Start>
</Response>

Example 4: Deepgram with Nova-3 model

<?xml version="1.0" encoding="UTF-8"?>
<Response>
  <Start>
    <Transcription
      transcriptionEngine="Deepgram"
      model="deepgram/nova-3"
      language="fr"
      transcriptionCallback="/transcription" />
  </Start>
</Response>

Example 5: Azure with Fast model and credentials

<?xml version="1.0" encoding="UTF-8"?>
<Response>
  <Start>
    <Transcription
      transcriptionEngine="Azure"
      model="azure/fast"
      language="en-US"
      apiKeyRef="your-azure-api-key-ref"
      region="eastus"
      transcriptionCallback="/transcription" />
  </Start>
</Response>

Example 6: Azure with Realtime model

<?xml version="1.0" encoding="UTF-8"?>
<Response>
  <Start>
    <Transcription
      transcriptionEngine="Azure"
      model="azure/realtime"
      language="en-US"
      region="westus"
      transcriptionCallback="/transcription" />
  </Start>
</Response>

Expected callbacks

If transcriptionCallback is set, transcription results are sent as they become available. See Transcription Callback for the full payload reference.

Guides

Markup Language

API Reference

For AI Agents

Attributes

Examples

Expected callbacks

​Attributes

​Examples

​Expected callbacks

Attributes

Examples

Expected callbacks