- Category
- AI Tools
- Rank
- No. 1507Tools index
Previous survey · No. 1514 ·
- Pricing
- Open Source
- Type
- APP
- Builder
- badlogic
- GitHub
- 120 stars
- Date
About
Voice recording and transcription app written in C.
What it does
Hold a system-wide hotkey, speak, then release it. Yakety records microphone audio, transcribes speech locally with a Whisper model, and pastes the resulting text into whichever application currently has focus.
Why it's ranked here
Yakety has a focused, practical workflow: dictation reaches the active application without a separate copy step or remote transcription service. Configurable hotkeys, model and language selection, voice activity detection, and model fallback add useful control. Platform constraints and macOS permission complexity narrow its audience.
What's good
Transcription stays local through Whisper integration. The application provides recording and transcription overlays, ignores extremely brief captures, cancels recording when another key interrupts it, and reports when no speech is detected. Users can change models, languages, hotkeys, voice activity detection, and launch-at-login behavior from the tray menu.
Tradeoffs
macOS support requires version 14 or later, Apple Silicon, microphone access, Accessibility access, and Input Monitoring. Command-line permissions inherit from the parent terminal application, which can make setup confusing. Windows builds require Visual Studio or Ninja, with Vulkan optional. The supplied workflow builds from source rather than documenting a package-manager installation.
How to use it well
Use Yakety for frequent short dictation directly into editors, chat clients, forms, or other focused applications. Choose a suitable local Whisper model, configure a comfortable hold-to-record shortcut, and run the tray application at login. It does not replace a meeting recorder, transcript archive, collaborative editor, or hosted transcription API.
Technical notes+
src/main.c coordinates model loading, keyboard callbacks, buffered recording, transcription, clipboard paste, overlays, signals, and cleanup. src/audio.c uses miniaudio to capture float32 mono audio at 16 kHz, expanding an in-memory buffer as needed. src/models.c calls transcription_init, removes a failed user-selected model, clears its preference, and retries with the bundled base model. src/menu.c exposes model, language, hotkey, VAD, and launch-at-login settings. src/recorder.c is a separate CLI recorder that writes Whisper-compatible WAV files. README.md documents CLI and app-bundle distribution through the build scripts.
Observed
- Primary language
- C, with a C++ transcription integration
- Interfaces
- System tray application and command-line tools
- Platforms
- macOS and Windows
- macOS support
- macOS 14.0 or later on Apple Silicon
- Transcription
- Local Whisper model processing through whisper.cpp
- Audio format
- 16 kHz mono float32 capture
- Build and packaging
- Source build scripts produce CLI tools and app bundles with embedded Whisper models
Read from README.md, src/main.c, src/app.h, src/http.h, src/menu.c, src/menu.h, src/audio.c, src/audio.h, src/utils.h, src/dialog.h, src/models.c, src/models.h, src/logging.h, src/overlay.h, src/recorder.c.
What it can do
Record voice audio
User's spoken words via microphone → Audio file
Transcribe audio to text
Audio recording → Text transcript
Save audio recordings
Voice recording session → Stored audio file
Save transcriptions
Transcribed text → Stored text file or document
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.