How stakira/openutau works
Project type and purpose
• OpenUtau is a full-stack, cross-platform desktop application for editing UTAU voice tracks.
• It uses a layered, component-based architecture: a rich Avalonia UI, a .NET Core “engine” library, a plugin system, and bundled native audio/ML libraries.
Top-level components
a. Presentation layer (UI)
• Technologies: Avalonia (XAML + C#), MVVM pattern
• Files: OpenUtau/*.axaml(.cs), ViewModels/, Views/, Controls/, Styles/, Strings/
• Responsibilities: render piano roll, dialogs, menus; capture user input; bind to ViewModels
b. Core engine library
• Project: OpenUtau.Core
• Sub-modules:
– Api: plugin interfaces (IG2p, IRenderer, IPlugin, PhonemizerRunner)
– Format: file I/O (USTx, VSQX, MusicXML, Wave)
– Analysis & Phonemizers: built-in dictionaries, Crepe ML model, DiffSinger, Enunu, Vogen, Voicevox adapters
– Classic: UTAU “classic” resampler runner, flag parser, voicebank loader
– Render: RenderEngine, RenderCache, Worldline integration
– SignalChain: mixing, faders, wave sources
– Audio: abstractions (IAudioOutput), backends (NAudio, MiniAudio)
– Util: file/INI/YAML parsers, path management, OS checks
• Responsibilities: load projects & voicebanks, run phonemizers, drive rendering, publish audio data to UI or files
c. Native modules (C++ worldline and dependencies)
• Directory: cpp/worldline + third_party
• Build system: Bazel, produces libworldline and friends
• Runtimes/: packaged native binaries (.dll/.so/.dylib) per OS/arch
• Responsibilities: low-level DSP (WORLDLINE resynthesis, fundamental frequency estimators, effects)
d. Plugin layer
• Project: OpenUtau.Plugin.Builtin (+ external plugins loaded via Core.Api)
• Built-in phonemizers for many languages and algorithms (VCV, CVVC, Arpasing, etc.)
• Responsibilities: extend or replace Core phonemizers; loaded at runtime by PluginLoader
e. Auxiliary scripts (Python)
• Directory: py/g2p
• Responsibilities: training G2P models; not part of the shipped runtime but important for contributors
f. Test suites
• Project: OpenUtau.Test
• Responsibilities: unit/integration tests for Core format parsers, plugins, worldline, UI tests
Relationships and data flows
• User → UI (Views & Controls) → ViewModels → Core.Api calls
• Core loads USTx/VSQX → Format module → Domain objects (UPart, UNote, etc.)
• Core invokes PluginRunner for phonemization → returns phoneme sequences
• Core.RenderEngine (Worldline wrapper) + SignalChain → raw audio buffers
• Audio module → platform-specific audio output → speakers
• UI polls PlaybackManager for progress and waveform previews
• PluginLoader uses reflection to discover IPlugin implementations in Plugin.Builtin and external .dlls
• Core library P/Invokes into native runtimes/libworldline
• File system stores voicebanks, settings, caches, render output
Architectural patterns & design principles
• MVVM for UI separation
• Layered architecture: Presentation → Application (ViewModels) → Domain/Engine → Infrastructure (native libs, file I/O)
• Plugin architecture following “inversion of control”: Core.Api interfaces, dynamic loading
• Single Responsibility: each Core subfolder focuses on one domain (Format, Phonemizers, Render, Audio)
• Cross-platform abstraction: IAudioOutput, LibraryLoader for native binaries