Deepgram Transcribe, Analyze, and Synthesize

arazzo: 1.0.1
info:
  title: Deepgram Transcribe, Analyze, and Synthesize
  summary: Transcribe audio to text, run text intelligence on the transcript, then synthesize a spoken response.
  description: >-
    An end-to-end voice round-trip that chains all three Deepgram speech AI
    surfaces. The workflow transcribes a pre-recorded audio file, runs text
    intelligence (summarization, sentiment, topics, intents) over the resulting
    transcript, and finally converts a chosen piece of text back into spoken
    audio with the Aura text-to-speech model. 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: speechToTextApi
  url: ../openapi/deepgram-speech-to-text-openapi.yml
  type: openapi
- name: textToSpeechApi
  url: ../openapi/deepgram-text-to-speech-openapi.yml
  type: openapi
workflows:
- workflowId: transcribe-analyze-synthesize
  summary: Transcribe audio, analyze the transcript text, and synthesize speech.
  description: >-
    Transcribes a hosted audio file, sends the transcript through text
    intelligence for a summary and sentiment, then synthesizes the summary back
    into speech audio.
  inputs:
    type: object
    required:
    - apiKey
    - audioUrl
    properties:
      apiKey:
        type: string
        description: Deepgram API key used to authenticate all requests.
      audioUrl:
        type: string
        description: Publicly accessible URL of the audio file to transcribe.
      sttModel:
        type: string
        description: Speech-to-text model to use for transcription.
        default: nova-3
      ttsModel:
        type: string
        description: Text-to-speech Aura voice to synthesize the response with.
        default: aura-asteria-en
  steps:
  - stepId: transcribeAudio
    description: >-
      Transcribe the hosted audio file with punctuation and smart formatting so
      the transcript is ready for text intelligence.
    operationId: transcribePreRecordedAudio
    parameters:
    - name: Authorization
      in: header
      value: Token $inputs.apiKey
    - name: model
      in: query
      value: $inputs.sttModel
    - 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
      requestId: $response.body#/metadata/request_id
  - stepId: analyzeTranscript
    description: >-
      Run text intelligence over the transcript to produce a summary along with
      sentiment, topics, and intents.
    operationId: analyzeText
    parameters:
    - name: Authorization
      in: header
      value: Token $inputs.apiKey
    - name: summarize
      in: query
      value: "true"
    - name: sentiment
      in: query
      value: true
    - name: topics
      in: query
      value: true
    - name: intents
      in: query
      value: true
    requestBody:
      contentType: application/json
      payload:
        text: $steps.transcribeAudio.outputs.transcript
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      summary: $response.body#/results/summary/short
      averageSentiment: $response.body#/results/sentiments/average/sentiment
  - stepId: synthesizeSummary
    description: >-
      Convert the generated summary text back into natural-sounding speech audio
      using the selected Aura voice.
    operationId: synthesizeSpeech
    parameters:
    - name: Authorization
      in: header
      value: Token $inputs.apiKey
    - name: model
      in: query
      value: $inputs.ttsModel
    - name: encoding
      in: query
      value: mp3
    requestBody:
      contentType: application/json
      payload:
        text: $steps.analyzeTranscript.outputs.summary
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      audioRequestId: $response.headers.x-request-id
  outputs:
    transcript: $steps.transcribeAudio.outputs.transcript
    summary: $steps.analyzeTranscript.outputs.summary
    averageSentiment: $steps.analyzeTranscript.outputs.averageSentiment