Home

Trackdub Engineering Wiki

Cross-platform, local-first AI dubbing workstation for Windows, macOS, and Linux.

Status: Pre-alpha · Stack: .NET 10, Avalonia, ONNX Runtime, FFmpeg

What is Trackdub?

Trackdub is a desktop dubbing pipeline: ingest media, transcribe, diarize, translate, synthesize speech, mix, and export. AI runs locally on your hardware.

Core principles:

• No fake readiness — provider registered, model downloaded, license verified, and stage succeeded are separate states
• Strict layer boundaries — Domain depends on nothing; UI never owns persistence or inference
• Manifest-driven models — every shipping model has a verified manifest entry
• Artifact preservation — skip/failure never silently overwrites usable artifacts

Quick start

git clone --recurse-submodules https://github.com/trackdubllc/Trackdub-gated.git
cd Trackdub-gated
dotnet build Trackdub.slnx -m:1
dotnet run --project src/Trackdub.App.Avalonia -f net10.0   # Linux/macOS
# Windows TFM: -f net10.0-windows10.0.19041.0

See Getting-Started for prerequisites, tests, CLI, and portal notes.

Wiki map

Section	Start here
Overview	Introduction-to-Trackdub
Architecture	Architecture-Overview → Layered-Dependency-Rules
Pipeline	Pipeline-Orchestration
Desktop App	Desktop-UI-Architecture
AI & Inference	Inference-Routing-and-Planning → Model-Manifests-and-Governance
Media & Audio	FFmpeg-Integration
Data	Local-Project-Persistence
Cloud & SaaS	Cloud-API-and-Job-Management
Development	CLI-and-Developer-Tools → Benchmarking-Framework

Pipeline at a glance

flowchart LR
    Ingest[Media Ingest] --> Prep[Audio Prep]
    Prep --> Sep[Speech Separation]
    Sep --> VAD[VAD]
    VAD --> Diar[Diarization]
    Diar --> ASR[Transcription]
    ASR --> Trans[Translation]
    Trans --> TTS[TTS]
    TTS --> Mix[Preview Mix]
    Mix --> Export[Export]

Loading

Optional stages (separation, lip sync) depend on project settings and model readiness. Stages are user-triggered from the desktop app, Sdk, or CLI.

Dependency graph (simplified)

App.Avalonia   → Application, Composition, Domain, Licensing, Media.Playback, Sdk
Cli            → Sdk
Sdk            → Application, Composition, Licensing
Application    → Contracts, Domain, Licensing
Infrastructure → Application, Contracts, Domain
Inference.Onnx → Inference, Contracts, Domain
Domain         → (none)

Full rules: Layered-Dependency-Rules.

Key docs

Doc	Purpose
AGENTS.md	Build/test/run commands
AGENT_CONTEXT.md	Product operating contract
AGENTS.md (pipeline/Sdk)	Dependency graph and headless commands
docs/	Architecture, decisions, specs

---

Use the sidebar for full navigation. Pages include prev/next links at the top.