StarWhisper - offline system-wide dictation for Windows/macOS powered by Whisper

[starwhisperai]

Sharing a desktop app built on Whisper (via the whisper.cpp runtime), in case it is useful to anyone who wants Whisper as their day-to-day dictation engine.

StarWhisper (https://starwhisper.ai): press a global hotkey anywhere in Windows or macOS, speak, and the transcript is typed into the focused application - editor, browser, chat, IDE. All inference runs locally on the user's machine (CPU, CUDA, Vulkan, or Metal), so no audio leaves the device.

• Models tiny through large-v3, chosen automatically from the machine's hardware
• Works in 100+ languages with auto-detect
• Free tier to try (500 words/day, no account), paid tier for unlimited

The interesting engineering has been everything around the model rather than the model itself: silence gating so short noise clips do not produce hallucinated text, filtering trailing-silence artifacts without ever destroying real speech, and hardware detection across a very long tail of consumer GPUs.

Happy to answer questions about any of it. Disclosure: I am the developer.