Deepgram Diarized Transcript Intelligence

arazzo: 1.0.1
info:
  title: Deepgram Diarized Transcript Intelligence
  summary: Transcribe a multi-speaker recording with diarization, then run targeted text intelligence over the transcript.
  description: >-
    A conversation analytics flow for multi-speaker recordings. The workflow
    transcribes a hosted audio file with diarization and utterance segmentation
    so each speaker turn is captured, then runs text intelligence over the
    transcript with summarization, sentiment, intents, and a set of custom
    topics supplied by the caller. Every step spells out its request inline so
    the flow can be read and executed without opening the underlying OpenAPI
    description.
  version: 1.0.0
sourceDescriptions:
- name: preRecordedApi
  url: ../openapi/deepgram-pre-recorded-api-openapi.yml
  type: openapi
- name: textIntelligenceApi
  url: ../openapi/deepgram-text-intelligence-api-openapi.yml
  type: openapi
workflows:
- workflowId: diarized-transcript-intelligence
  summary: Diarize a recording, then analyze the transcript for summary, sentiment, and custom topics.
  description: >-
    Transcribes a multi-speaker recording with diarization and utterances, then
    sends the transcript through text intelligence scoped to caller-supplied
    custom topics.
  inputs:
    type: object
    required:
    - apiKey
    - audioUrl
    - customTopics
    properties:
      apiKey:
        type: string
        description: Deepgram API key used to authenticate requests.
      audioUrl:
        type: string
        description: Publicly accessible URL of the multi-speaker audio file.
      model:
        type: string
        description: Speech-to-text model to use for transcription.
        default: nova-3
      customTopics:
        type: array
        description: Custom topics to detect within the transcript text.
        items:
          type: string
  steps:
  - stepId: transcribeDiarized
    description: >-
      Transcribe the recording with diarization, utterances, punctuation, and
      smart formatting so each speaker turn is captured.
    operationId: transcribePreRecordedAudio
    parameters:
    - name: Authorization
      in: header
      value: Token $inputs.apiKey
    - name: model
      in: query
      value: $inputs.model
    - name: diarize
      in: query
      value: true
    - name: utterances
      in: query
      value: true
    - name: punctuate
      in: query
      value: true
    - name: smart_format
      in: query
      value: true
    requestBody:
      contentType: application/json
      payload:
        url: $inputs.audioUrl
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      transcript: $response.body#/results/channels/0/alternatives/0/transcript
      utterances: $response.body#/results/utterances
      requestId: $response.body#/metadata/request_id
  - stepId: analyzeConversation
    description: >-
      Run text intelligence over the diarized transcript with summarization,
      sentiment, intents, and the caller-supplied custom topics.
    operationId: analyzeText
    parameters:
    - name: Authorization
      in: header
      value: Token $inputs.apiKey
    - name: summarize
      in: query
      value: "true"
    - name: sentiment
      in: query
      value: true
    - name: intents
      in: query
      value: true
    - name: custom_topic
      in: query
      value: $inputs.customTopics
    requestBody:
      contentType: application/json
      payload:
        text: $steps.transcribeDiarized.outputs.transcript
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      summary: $response.body#/results/summary/short
      averageSentiment: $response.body#/results/sentiments/average/sentiment
      topics: $response.body#/results/topics/segments
  outputs:
    transcript: $steps.transcribeDiarized.outputs.transcript
    utterances: $steps.transcribeDiarized.outputs.utterances
    summary: $steps.analyzeConversation.outputs.summary
    topics: $steps.analyzeConversation.outputs.topics