Introduction to Trackdub

Introduction to Trackdub

Section: Overview · Home · Getting-Started

Trackdub is a cross-platform, local-first AI dubbing workstation for Windows, macOS, and Linux. Built on .NET 10 and Avalonia, it runs media ingest, transcription, diarization, translation, TTS, mix, and export on the user's hardware with ONNX-backed models.

Architecture at a glance

Domain-Driven Design with a strict layered dependency model. Domain and Contracts are roots. The Avalonia shell sits above Application/Sdk and never owns inference or persistence.

graph TD
    App[App.Avalonia] --> Application
    App --> Composition
    App --> Domain
    App --> MediaPlayback[Media.Playback]
    App --> Sdk

    Cli[Trackdub.Cli] --> Sdk
    Sdk --> Composition
    Sdk --> Application

    Application --> Contracts
    Application --> Domain

    Infrastructure --> Application
    Infrastructure --> Contracts
    Infrastructure --> Domain

    Inference --> Contracts
    Inference --> Domain
    InferenceOnnx[Inference.Onnx] --> Inference

    Composition --> Application
    Composition --> Inference
    Composition --> Infrastructure
    Composition --> Media

Loading

Layer descriptions

Layer	Responsibility
Trackdub.App.Avalonia	Desktop UI shell
Trackdub.Domain	Entities and invariants; zero external dependencies
Trackdub.Contracts	DTOs and shared contracts
Trackdub.Application	Pipeline orchestration, stage planning, execution snapshots
Trackdub.Infrastructure	SQLite persistence, filesystem, logging
Trackdub.Inference	Model abstractions, manifests, EP selection
Trackdub.Inference.Onnx	ONNX Runtime / Windows ML implementations
Trackdub.Media	FFmpeg probing, extraction, mixing
Trackdub.Media.Playback	libmpv / LibVLC playback backends
Trackdub.Sdk / Cli	Headless automation and CLI

The dubbing pipeline

Stages are explicit, ordered, and resumable:

1. Media ingest / normalized audio
2. Optional speech separation
3. VAD → Diarization → ASR
4. Translation → TTS
5. Preview mix → Export

Stages are user-triggered. Loading media alone does not auto-run the full pipeline.

See Pipeline-Orchestration.

Model governance

Bundled inventory: src/Trackdub.Inference/Runtime/ModelManifest/bundled-models.manifest.json (under external/Trackdub/ in the app checkout).

• Commercial licenses only for shipping lanes
• Unknown license = unsafe
• Provider registered ≠ model downloaded ≠ stage ran ≠ stage succeeded

Build snapshot

git clone --recurse-submodules https://github.com/trackdubllc/Trackdub-gated.git
cd Trackdub-gated
dotnet build Trackdub.slnx -m:1

Full setup: Getting-Started. Architecture depth: Architecture-Overview · Layered-Dependency-Rules.