> This page is for APIs.

> Fetch clean Markdown by appending `.md` to any page URL under https://signalwire.com/docs or requesting it with the HTTP header `Accept: text/markdown`. The root index at https://signalwire.com/docs/llms.txt lists the available documentation indexes.

# ai.transcribe.utterance

GET /

Receive ai.transcribe.utterance events

Reference: https://signalwire.com/docs/apis/relay/client/fabric/ai-transcribe-utterance

## AsyncAPI Specification

```yaml
asyncapi: 2.6.0
info:
  title: ai.transcribe.utterance
  version: subpackage_aiTranscribeUtterance.ai.transcribe.utterance
  description: Receive ai.transcribe.utterance events
channels:
  /:
    description: Receive ai.transcribe.utterance events
    publish:
      operationId: subpackage_aiTranscribeUtterance.ai.transcribe.utterance-publish
      summary: ai.transcribe.utterance
      message:
        name: onAiTranscribeUtteranceFabricAiTranscribeUtteranceEvent
        title: ai.transcribe.utterance
        payload:
          $ref: >-
            #/components/schemas/ai.transcribe.utterance_fabricAiTranscribeUtteranceEvent
servers:
  production:
    url: wss://relay.signalwire.com/
    protocol: wss
    x-default: true
components:
  schemas:
    ChannelsAiTranscribeUtteranceMessagesFabricAiTranscribeUtteranceEventJsonrpc:
      type: string
      enum:
        - '2.0'
      description: JSON-RPC version. Always `2.0`.
      title: >-
        ChannelsAiTranscribeUtteranceMessagesFabricAiTranscribeUtteranceEventJsonrpc
    ChannelsAiTranscribeUtteranceMessagesFabricAiTranscribeUtteranceEventMethod:
      type: string
      enum:
        - signalwire.event
      description: Always `signalwire.event`.
      title: >-
        ChannelsAiTranscribeUtteranceMessagesFabricAiTranscribeUtteranceEventMethod
    ChannelsAiTranscribeUtteranceMessagesFabricAiTranscribeUtteranceEventParamsEventType:
      type: string
      enum:
        - ai.transcribe.utterance
      description: The event type — identifies which event this is.
      title: >-
        ChannelsAiTranscribeUtteranceMessagesFabricAiTranscribeUtteranceEventParamsEventType
    FabricFabricAiTranscribeUtteranceParamsConfidence:
      oneOf:
        - type: number
          format: double
        - description: Any type
      description: >-
        Recognition confidence for this utterance as reported by the engine,
        roughly 0.0-1.0. `null` when the engine reports none.
      title: FabricFabricAiTranscribeUtteranceParamsConfidence
    Calling.TranscribeUtterance:
      type: object
      properties:
        role:
          type: string
          description: >-
            Who spoke: `remote-caller` (the far end) or `local-caller` (your
            side).
        content:
          type: string
          description: The transcribed text.
        lang:
          type: string
          description: >-
            The language of the utterance as a BCP-47 language code (e.g. `en`,
            `es-ES`).
        utterance_id:
          type: string
          description: >-
            Unique identifier for this utterance, for correlation and
            de-duplication.
        timestamp:
          type: integer
          format: int64
          description: >-
            When the utterance was produced, as a Unix timestamp in
            microseconds.
        confidence:
          type: number
          format: double
          description: >-
            Recognition confidence for the utterance, roughly 0.0-1.0. Omitted
            when the engine reports none.
        tokens:
          type: integer
          description: Number of tokens in the utterance.
        char_count:
          type: integer
          description: Number of characters in the utterance text.
        word_count:
          type: integer
          description: Number of words in the utterance text.
        silence_before_ms:
          type: integer
          description: >-
            Milliseconds since this speaker's previous utterance. Omitted on the
            speaker's first utterance.
        filter_type:
          type: string
          description: >-
            The kind of content filter that matched this utterance, when one
            did.
        filter_value:
          type: string
          description: The value the content filter matched, when one did.
        start:
          type: number
          format: double
          description: >-
            Engine-reported start offset of the utterance, in seconds. Present
            with `verbose_utterances`.
        end:
          type: number
          format: double
          description: >-
            Engine-reported end offset of the utterance, in seconds. Present
            with `verbose_utterances`.
        duration:
          type: number
          format: double
          description: >-
            Engine-reported duration of the utterance, in seconds. Present with
            `verbose_utterances`.
        duration_ms:
          type: number
          format: double
          description: >-
            Duration of the utterance, in milliseconds. Present with
            `verbose_utterances`.
        words:
          type: array
          items:
            description: Any type
          description: >-
            Engine-reported word-level detail (timings, confidence). Present
            with `verbose_utterances`; the shape is engine-specific.
        alternatives:
          type: array
          items:
            description: Any type
          description: >-
            Engine-reported alternative transcripts. Present with
            `verbose_utterances`; the shape is engine-specific.
        request_id:
          type: string
          description: Engine request identifier. Present with `verbose_utterances`.
        detected_language:
          type: string
          description: Language detected by the engine. Present with `verbose_utterances`.
        speech_final:
          type: boolean
          description: >-
            Whether the engine marked this as the end of a spoken phrase.
            Present with `verbose_utterances`.
        is_final:
          type: boolean
          description: >-
            Whether the engine marked this result as final. Present with
            `verbose_utterances`.
      required:
        - role
        - content
        - lang
        - utterance_id
        - timestamp
      description: A single transcribed utterance from a live-transcription session.
      title: Calling.TranscribeUtterance
    Calling.AiCallInfo:
      type: object
      properties:
        project_id:
          type: string
          description: Your project ID.
        space_id:
          type: string
          description: Your space ID.
        content_type:
          type: string
          description: The content type of the delivery.
        content_disposition:
          type: string
          description: The content disposition of the delivery.
        conversation_type:
          type: string
          description: The kind of conversation this delivery describes.
        call_id:
          type: string
          description: The call ID.
      required:
        - content_type
        - content_disposition
        - conversation_type
        - call_id
      description: >-
        Call/session context attached to AI transcribe, translate, and sidecar
        webhook deliveries, as a

        sibling of the documented event fields. Added on every delivery.
      title: Calling.AiCallInfo
    Calling.AiChannelData:
      type: object
      properties:
        project_id:
          type: string
          description: Your project ID.
        space_id:
          type: string
          description: Your space ID.
        content_type:
          type: string
          description: The content type of the delivery.
        content_disposition:
          type: string
          description: The content disposition of the delivery.
        call_id:
          type: string
          description: The call ID.
        call_start_date:
          type: integer
          format: int64
          description: When the call started, as a Unix timestamp in microseconds.
        call_answer_date:
          type: integer
          format: int64
          description: >-
            When the call was answered, as a Unix timestamp in microseconds. `0`
            if not answered yet.
        call_end_date:
          type: integer
          format: int64
          description: >-
            When the call ended, as a Unix timestamp in microseconds. `0` while
            the call is still up.
        caller_id_name:
          type: string
          description: The caller's name.
        caller_id_number:
          type: string
          description: The caller's number.
        SWMLVars:
          type: object
          additionalProperties:
            description: Any type
          description: >-
            Custom variables set on the call during its SWML execution, as
            name/value pairs.
        SWMLCall:
          type: object
          additionalProperties:
            description: Any type
          description: >-
            The SWML `call` object for this call — the call metadata (direction,
            from/to, headers, timing) available to your SWML document. See the
            SWML reference for its fields.
      required:
        - content_type
        - content_disposition
        - call_id
      description: >-
        Extended call/channel context attached to AI transcribe and translate
        deliveries — a superset of

        `AiCallInfo` with call timing and caller-ID detail.
      title: Calling.AiChannelData
    Fabric.FabricAiTranscribeUtteranceParams:
      type: object
      properties:
        call_id:
          type: string
          description: >-
            Unique identifier of the call this event is about. Always present on
            Fabric events.
        node_id:
          type: string
          description: The node the call leg lives on.
        room_session_id:
          type: string
          description: >-
            The room session this call leg belongs to. In a 1:1 call this equals
            the `call_id`.
        segment_id:
          type: string
          description: >-
            Identifier of the call segment this event belongs to, for
            correlating events within the same segment.
        tag:
          type: string
          description: >-
            The `tag` the leg was dialed with, when a server-side application
            set one on the dial. Absent otherwise.
        confidence:
          $ref: >-
            #/components/schemas/FabricFabricAiTranscribeUtteranceParamsConfidence
          description: >-
            Recognition confidence for this utterance as reported by the engine,
            roughly 0.0-1.0. `null` when the engine reports none.
        utterance:
          $ref: '#/components/schemas/Calling.TranscribeUtterance'
          description: The recognized utterance.
        call_info:
          $ref: '#/components/schemas/Calling.AiCallInfo'
          description: Call/session context for this delivery.
        channel_data:
          $ref: '#/components/schemas/Calling.AiChannelData'
          description: Extended call/channel context for this delivery.
      required:
        - call_id
        - node_id
        - room_session_id
        - confidence
        - utterance
        - call_info
        - channel_data
      title: Fabric.FabricAiTranscribeUtteranceParams
    ChannelsAiTranscribeUtteranceMessagesFabricAiTranscribeUtteranceEventParams:
      type: object
      properties:
        event_type:
          $ref: >-
            #/components/schemas/ChannelsAiTranscribeUtteranceMessagesFabricAiTranscribeUtteranceEventParamsEventType
          description: The event type — identifies which event this is.
        event_channel:
          type: string
          description: >-
            The channel the event was delivered on — for Calling events, the
            `protocol`

            string your connection received from `signalwire.connect`.
        timestamp:
          type: number
          format: double
          description: >-
            When the event was emitted, as a Unix timestamp in seconds. Always
            present.
        project_id:
          type: string
          description: >-
            Your project ID. Present on Calling events and most Call Fabric
            events; absent on `call.joined`, `call.left`, `member.joined`, and
            `member.left`.
        space_id:
          type: string
          description: >-
            Your space ID. Present on Calling events and most Call Fabric
            events; absent on `call.joined`, `call.left`, `member.joined`, and
            `member.left`.
        params:
          $ref: '#/components/schemas/Fabric.FabricAiTranscribeUtteranceParams'
          description: The event-specific payload.
      required:
        - event_type
        - event_channel
        - timestamp
        - params
      description: The event envelope.
      title: >-
        ChannelsAiTranscribeUtteranceMessagesFabricAiTranscribeUtteranceEventParams
    ai.transcribe.utterance_fabricAiTranscribeUtteranceEvent:
      type: object
      properties:
        jsonrpc:
          $ref: >-
            #/components/schemas/ChannelsAiTranscribeUtteranceMessagesFabricAiTranscribeUtteranceEventJsonrpc
          description: JSON-RPC version. Always `2.0`.
        id:
          type: string
          description: >-
            Event id — a unique UUID generated for each event. Events are
            notifications: you do not reply to them.
        method:
          $ref: >-
            #/components/schemas/ChannelsAiTranscribeUtteranceMessagesFabricAiTranscribeUtteranceEventMethod
          description: Always `signalwire.event`.
        params:
          $ref: >-
            #/components/schemas/ChannelsAiTranscribeUtteranceMessagesFabricAiTranscribeUtteranceEventParams
          description: The event envelope.
      required:
        - jsonrpc
        - id
        - method
        - params
      description: >-
        Fires on a Call Fabric call leg when live transcription produces an
        utterance. Same fields as the Calling API's
        `calling.ai.transcribe.utterance` event. SWML or a server-side
        application starts live transcription on the leg; there is no client
        request that starts it.
      title: ai.transcribe.utterance_fabricAiTranscribeUtteranceEvent

```