All 11 Standard Workflow Steps

DuRT 3.0’s Workflow Engine features an extensible, DAG-based pipeline architecture with 11 specialized step types. Each step adheres to a strongly-typed data contract, enabling you to chain and branch inputs, recognition engines, language models, voiceover, and multi-cloud delivery into an automated pipeline.

┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│ 1. Input Source │ ──> │ 2. ASR & Stream │ ──> │ 3. AI Post-Proc │ ──> │ 4. TTS & Export │
└─────────────────┘     └─────────────────┘     └─────────────────┘     └─────────────────┘

---

Group 1: Input Steps (Input Sources)

1. monitorCapture — System Audio & Screen Recording

• Role: Continuously captures live audio from Safari, Zoom, YouTube, or your microphone via macOS ScreenCaptureKit and CoreAudio.
• Output Port Type: audioStream (PCM audio stream)
• Key Options: System Internal Audio vs. Microphone; hardware/DSP Acoustic Echo Cancellation (AEC).

2. mediaInput — File & Streaming URL Ingestion

• Role: Ingests local media files (.mp4, .mov, .mkv, .mp3, .m4a, .wav) or streaming web links (YouTube, Bilibili, direct media URLs).
• Output Port Type: mediaFile (Physical media asset)
• Key Options: Client-side audio demuxing and format standardization.

3. subtitleInput — Subtitle & Raw Text Ingestion

• Role: Ingests pre-existing .srt or .vtt subtitle files or references text from prior tasks, allowing you to run downstream AI without re-running speech recognition.
• Output Port Type: subtitleTrack (Timed subtitle track)
• Key Options: Automatic text encoding detection and timeline timecode offset shifting.

---

Group 2: Speech Recognition (ASR) & Live Translation

4. realtimeASR — Streaming Speech Recognition

• Role: Emits real-time subtitle deltas and finalized sentences with sub-second latency to drive floating captions.
• Input Port Type: audioStream
• Output Port Type: subtitleStream (Live subtitle stream)
• Supported Engines: Apple On-Device Speech, OneASR WebSocket Gateway, OpenAI Realtime API.

5. mediaASR — File Transcription & Speaker Diarization

• Role: High-precision offline transcription for long audio files, generating millisecond word-level timestamps and unsupervised speaker diarization (Speaker 1, Speaker 2, etc.).
• Input Port Type: mediaFile
• Output Port Type: subtitleTrack (Timed subtitle track)
• Supported Engines: Apple On-Device Speech, OneASR Gateway (Whisper / SenseVoice), OpenAI Whisper API.

6. realtimeTranslation — Simultaneous Interpretation

• Role: Translates streaming ASR segments on the fly, sentence by sentence, for live bilingual meeting captions.
• Input Port Type: subtitleStream
• Output Port Type: subtitleStream (Bilingual live stream)
• Engines: Apple On-Device Neural Translation (macOS 15), DeepSeek API, OpenAI.

---

Group 3: AI Post-Processing, Synthesis & Delivery

7. subtitleRefinement — Subtitle Formatting & Refinement

• Role: AI-driven cleanup that removes filler words ("um", "like"), corrects punctuation, and optimizes line breaks (5-8 seconds per subtitle card) while preserving original timestamps.
• Input Port Type: subtitleTrack
• Output Port Type: subtitleTrack (Refined subtitle track)

8. fullTranslation — Full Document & Subtitle Translation

• Role: Context-aware batch translation preserving exact subtitle index codes and timecodes across 50+ languages.
• Input Port Type: subtitleTrack
• Output Port Type: subtitleTrack (Bilingual subtitle track)

9. contentCreation — Summary & Content Derivation

• Role: Automatically extracts executive summaries, meeting minutes, action items, or social media show notes from transcripts.
• Input Port Type: subtitleTrack or document
• Output Port Type: document (Markdown / TXT document)

10. textToSpeech — Voice Synthesis & Dubbing

• Role: Converts polished or translated subtitle lines into audio tracks using realistic neural voices (OpenAI TTS, Xiaomi MIMO TTS, OneTTS).
• Input Port Type: subtitleTrack
• Output Port Type: mediaFile (Synthesized audio file)

11. exportArtifact — Delivery & Multi-Cloud Sync

• Role: Exports finalized assets to local storage and triggers automated delivery across 6 cloud channels:
  • Local export: .srt, .vtt, .json, .txt, .md, .mp3
  • Automated cloud delivery: S3-compatible Object Storage (AWS S3, Cloudflare R2, MinIO, OSS), WebDAV Drives (Nextcloud, Synology), Local / NAS Folders, Notion Databases, Obsidian Vaults, and Custom HTTP Webhooks.
• Input Port Type: subtitleTrack, document, mediaFile
• Output Port Type: deliverable (Execution summary receipt)