Salad Speech-to-Text

naftiko: "1.0.0-alpha1"

info:
  label: "Salad Speech-to-Text"
  description: >-
    Workflow capability for speech-to-text transcription using Salad's distributed
    GPU inference network. Supports multi-language transcription, speaker diarization,
    word-level timestamps, and SRT caption generation for audio and video content.
  tags:
    - Audio Transcription
    - Captions
    - Diarization
    - Media Processing
    - Salad
    - Speech Recognition
    - Subtitles
    - Video Transcription
  created: "2026-05-02"
  modified: "2026-05-02"

binds:
  - namespace: env
    keys:
      SALAD_API_KEY: SALAD_API_KEY
      SALAD_API_URL: SALAD_API_URL

capability:
  consumes:
    - import: salad-transcription
      location: ./shared/salad-transcription-api.yaml

  exposes:
    - type: rest
      port: 8080
      namespace: salad-speech-to-text-api
      description: "Unified REST API for Salad speech-to-text transcription workflows."
      resources:
        - path: /v1/transcriptions
          name: transcriptions
          description: "Submit audio and video files for transcription."
          operations:
            - method: POST
              name: transcribe-media
              description: "Submit a media URL for transcription with language and output options."
              call: "salad-transcription.transcribe-media"
              outputParameters:
                - type: object
                  mapping: "$."
        - path: /v1/transcriptions/{jobId}
          name: transcription-by-id
          description: "Retrieve transcription results for a completed job."
          operations:
            - method: GET
              name: get-transcript
              description: "Retrieve the full transcript, segments, and optional SRT output."
              call: "salad-transcription.get-transcript"
              with:
                jobId: "rest.jobId"
              outputParameters:
                - type: object
                  mapping: "$."

    - type: mcp
      port: 9080
      namespace: salad-speech-to-text-mcp
      transport: http
      description: "MCP server for AI-assisted Salad speech-to-text transcription."
      tools:
        - name: transcribe-audio-video
          description: >-
            Submit an audio or video file URL to Salad for speech-to-text transcription.
            Supports 97 languages, speaker diarization, word-level timestamps, and SRT output.
            Returns a job ID to retrieve results.
          hints:
            readOnly: false
            destructive: false
          call: "salad-transcription.transcribe-media"
          outputParameters:
            - type: object
              mapping: "$."
        - name: get-transcription-result
          description: >-
            Retrieve the completed transcription for a job by ID. Returns segments,
            word timestamps, speaker labels, and optional SRT caption content.
          hints:
            readOnly: true
            idempotent: true
          call: "salad-transcription.get-transcript"
          with:
            jobId: "tools.jobId"
          outputParameters:
            - type: object
              mapping: "$."