Skip to content

ROAR v0.24.0 — Read Aloud (local, GPU-accelerated speech)

Choose a tag to compare

View all tags
@xhan145 xhan145 released this 25 Jul 03:20
· 114 commits to main since this release
v0.24.0
1aefc86

Read Aloud comes to ROAR. ROAR can now speak — select or type text and have it read back in a natural local voice (Kokoro-82M), entirely on your own machine. No cloud, no account, nothing leaves your computer.

New: Read Aloud (Kokoro TTS)

  • Fully local & private — synthesis runs in an isolated worker; ROAR never sends your text or audio anywhere. Read Aloud is off until you enable it, and Core dictation is unchanged.
  • Optional download — to keep the base install lean, the voice model and its runtime are a one-time setup from Settings → Read Aloud, not bundled in this installer.
  • GPU accelerated — with an NVIDIA GPU, setup installs CUDA and speech is ~10× faster (about a third of a second per sentence on an RTX 4060). CPU works everywhere. The model is preloaded and kept warm, and a one-time warm-up keeps the first request quick.

Everything else

Same fast, private local dictation as v0.23.0 — CUDA/Vulkan GPU paths, snippets, speech cleanup, per-app profiles, offline licensing. Your settings, history, and licence carry over.

Install

Run the setup and follow the prompts. Requires: Windows 10/11.

The model + runtime for Read Aloud are set up from inside the app after install; the GPU download needs a few GB of free disk.

Full detail: CHANGELOG.md

Assets 3
Loading