Ready to dictate
- Speech model
- Microphone
- Hotkey
Modellarge-v3-turbo · GPU
Start with Windows
Show notifications
Under the hood
A native core, not a wrapper
SweetWhisper is not an Electron shell around a cloud API. It is a native Rust application that runs neural networks right on your GPU — any GPU: NVIDIA, AMD or Intel.
- Engines
- Whisper (GGML), Parakeet and GigaAM (ONNX) — switch in two clicks
- GPU
- Vulkan + DirectML: runs on NVIDIA, AMD and Intel with no CUDA dependency
- Core
- Rust + Tauri: a single ~90 MB app, no background services
- Accuracy
- Your own correction dictionary, plus a filter that drops the phantom phrases Whisper sometimes invents in silence
- LLM post-processing
- Optional text post-processing with a local model — offline too
- Open models
- All models are open (MIT / Apache-2.0 / CC-BY), licenses listed in the app
Interface
One window instead of a maze of settings
Model, speed, hotkeys and text insertion — all in one place. It checks your microphone, model and hotkey for you and tells you when you’re ready to dictate.
↗ Works offlineGPU · Vulkan / DirectML / CUDA
- Dictation readiness indicator
- Model catalog in one click
- Dark and light themes
- History and stats — local only