Skip to content

Diarix v0.1.0-alpha.1: Windows Core

Latest
Latest

Choose a tag to compare

View all tags
@Yannam-Builds Yannam-Builds released this 20 Jul 00:00
v0.1.0-alpha.1
a3f9948

Diarix v0.1.0-alpha.1: Windows Core

This is the first public Windows 11 x64 pre-alpha of Diarix, a transcription-first local speech studio.

Install

  1. Download Diarix.Core.Setup.0.1.0-alpha.1.exe from the assets below.
  2. Run the installer. If Windows SmartScreen appears, choose More info, then Run anyway.
  3. Open Diarix. The compact CPU runtime, FFmpeg, and FFprobe are included.
  4. Download models or the optional CUDA backend from inside the app when you need them.

The installer checks Microsoft WebView2 and the Visual C++ runtime. If either dependency is missing, it opens the official Microsoft download page.

Model weights are not bundled. Choose the model and cache location in Diarix, then use the Models section to download only what you need.

Which asset should I download?

Most users need only Diarix.Core.Setup.0.1.0-alpha.1.exe.

The diarix-server-cuda and cuda-libs archives are backend components consumed by Diarix. Do not extract or install them manually. The app downloads, verifies, and assembles them from GPU settings.

The .sha256 files are optional checksums for manual verification.

Included in Core

  • Native Diarix desktop app
  • Compact local CPU server with transcribe.cpp
  • FFmpeg and FFprobe media ingestion
  • Dependency checks for WebView2 and the Visual C++ runtime
  • Empty model cache stored in the location you choose

Verified in this build

  • Installed-app startup without a development server or terminal window
  • Compact CPU sidecar health and clean process ownership
  • Packaged CUDA runtime on an NVIDIA RTX 5070 Ti
  • Runtime imports for Whisper, WhisperX, Faster-Whisper, NeMo and Canary, Qwen3-ASR, transcribe.cpp, local TTS engines, and Qwen refinement
  • 188 backend tests passed, with 2 intentionally skipped
  • Frontend typecheck and production build passed

Requirements and known limits

  • Windows 11 x64
  • Approximately 1 GB free for Core, plus storage for models or CUDA
  • NVIDIA GPU optional; CPU transcription is included
  • Unsigned pre-alpha, so Windows may display a SmartScreen warning
  • IBM Granite Speech 3.3 8B was excluded from this GPU verification pass
  • macOS and Linux are not supported by this release

Please report reproducible problems through GitHub Issues.

Thanks

Diarix builds on the Voicebox studio foundation. Its lightweight push-to-talk direction and warm model lifecycle were strongly informed by Handy. Thank you to both projects and their contributors.

Assets 11
Loading