Logo Lanfrica

kenikiara/bonga

Record type:

softwaretools
Creator:
ken
Host:
BONGA — free voice typing for Windows. Speak, it types. Made in Kenya. # BONGA A native Windows voice-dictation app modeled on Wispr Flow: hold a hotkey, speak, release — and clean, polished text appears in whatever app you're typing in. Works in Gmail, Slack, VS Code, terminals, anywhere text goes. Unlike Wispr Flow (cloud-only), BONGA transcribes **entirely on-device** with Whisper (whisper.cpp via Whisper.net). No account, no API key, no audio ever leaves your PC. An optional bring-your-own-key cloud mode is included. ## Features - **Push-to-talk dictation** — hold `Right Ctrl` (configurable), speak, release. Text is inserted into the focused app. - **Hands-free mode** — quick-tap the hotkey to start, press again to finish. `Esc` cancels. - **Flow Bar** — floating pill at the bottom of the screen with live waveform while recording; click it to dictate. Never steals focus. - **AI auto-edits** — removes filler words (um, uh…), collapses stutters, auto-capitalizes, punctuates. Voice commands: say *"new line"* / *"new paragraph"*. - **Personal dictionary** — add names/jargon (biases the recognizer via decode prompt) and hard corrections (*misheard → correct*). - **Snippets** — say a trigger phrase ("insert my email") to expand full text blocks. - **History & stats** — past dictations with copy buttons, words written, time saved vs typing. - **100+ languages** — auto-detect or pin a language (Whisper multilingual models). - **Tray app** — runs in the background; launch-at-startup option. - **Optional cloud** — point at any OpenAI-compatible endpoint for cloud STT and/or an LLM "polish" pass. ## Build & run Requires the .NET 8 SDK. ```powershell dotnet build src/VoiceFlow/VoiceFlow.csproj -c Release dotnet run --project src/VoiceFlow -c Release ``` Self-contained distributable (no .NET install needed on the target machine): ```powershell dotnet publish src/VoiceFlow/VoiceFlow.csproj -c Release -r win-x64 --self-contained -o dist ``` On first run, download the speech model from the Home screen (base-q5_1, ~60 MB, one time). Model …