Local Whisper dictation for Windows with GPU-accelerated transcription
VoZii transcribes speech locally using OpenAI's Whisper via whisper.cpp, inserting text at the cursor in any Windows application. haZii.org publishes this privacy-minded Windows utility as a portable executable. Key functions include GPU acceleration (CUDA and Vulkan), global push-to-talk hotkeys, and optional Ollama post-processing for grammar and formatting. The tool fits privacy-conscious writers, developers, and users with accessibility needs who prefer local model processing and direct text insertion over cloud services.
What does VoZii actually do inside the desktop workflow?
VoZii runs Whisper models locally through whisper.cpp and types transcribed text into the active window at the cursor position, so dictated text appears in editors, browsers, or chat apps without switching context. The app is delivered as a single portable 64-bit executable and offers a global hotkey in push-to-talk or toggle modes to control when audio is captured.
Does VoZii affect system performance during transcription?
The app can run on CPU-only systems but is specifically designed to use GPU acceleration via CUDA for NVIDIA or Vulkan for AMD and Intel, which reduces transcription latency compared with CPU processing. Because it uses local hardware for model inference, the measurable impact depends on whether the system routes model inference to the GPU or CPU; GPU-equipped machines will see faster, near-real-time typing of text.
Is VoZii safe for sensitive audio and appropriate for regular workflows?
VoZii performs recognition locally after an initial model download, so no audio is sent to external servers during normal use. Optional integration with Ollama runs post-processing on a local model, keeping data on the machine. The app's portable design avoids installation changes to the system, which helps when you need a tool that does not write persistent entries across multiple machines.
A strong local dictation tool for technically minded, privacy-focused users
VoZii suits users who accept the responsibility of managing local models and hardware acceleration to gain on-device transcription and privacy. Expect some setup work when enabling GPU acceleration or integrating a local Ollama instance; users who prefer a plug-and-play, cloud-powered dictation service may find that trade-off unwelcome. Recommended for anyone who prioritizes local processing and direct text insertion in desktop workflows.
