Start

The <Start> verb is used to begin media operations during a call. This includes starting a recording session or transcription service for the call leg. It can be inserted inside a <Dial> block to start these operations on the connected call, or independently to start live services before a connection.

You can use multiple <Start> verbs to begin different operations in parallel.

<Start media="recording|transcription"
       direction="inbound|outbound|both"
       format="wav|mp3"
       channels="1|2"
       sample_rate="8000|16000"
       transcript_url="url" />

---

Attribute	Type	Required	Default	Description
media	string	✔️	—	Type of media to start: recording or transcription.
direction	string	❌	both	Audio direction: inbound, outbound, or both.
format	string	❌	wav	Recording format (wav, mp3, etc.).
channels	integer	❌	1	Number of audio channels: 1 (mono) or 2 (stereo).
sample_rate	integer	❌	8000	Audio sample rate in Hz.
transcript_url	URI	❌	—	URL to receive transcription webhooks. Only used when media="transcription".

---

<Response>
    <Dial caller="18005551234">
        <Connect to="+15556667777" />
        <Start media="recording" format="mp3" channels="2" sample_rate="16000" />
        <Start media="transcription" transcript_url="https://example.com/api/transcripts" />
    </Dial>
</Response>

• When media="recording", audio is captured according to the provided parameters (format, direction, etc.).
• When media="transcription", the call's audio is streamed to a speech-to-text engine and results are POSTed to the transcript_url.
• A <Start> verb should follow a successful <Connect> if targeting the bridged call.
• Supports multiple <Start> verbs within a <Dial>.

{
"recording_url": "https://.../recording.mp3",
"duration": 84,
"format": "mp3",
"start_time": "2025-07-17T15:00:00Z",
"end_time": "2025-07-17T15:01:24Z"
}

• transcription_started
• transcription_partial
• transcription_completed

• Invalid media types are ignored.
• If transcript_url is missing for transcription, events are not delivered.
• Errors are logged silently to avoid interrupting the call flow.

• Use after a <Connect> verb to capture bridged audio.
• Supports layered actions — for example, record and transcribe the same call.
• Do not rely on transcription accuracy for compliance use cases.