Turn dictation from sequential to parallel: speak while you trigger, instead of triggering before you speak.
-87% bin size, -83% ram (780mb → 132mb). vs Wispr Flow.
Local & cloud STT solution both available.
Most dictation apps make you reach for a hotkey and wait before speaking. Roma Just Talk keeps a short rolling buffer, so you can begin speaking while your hand moves to Shift without losing the start of the thought.
- Speak before the hotkey: a short pre-roll buffer can include the words you said just before triggering capture
- Pre-roll capture UX: not always transcribing, just buffering enough that recording does not start from zero
- Local or cloud: use local models when you want control, cloud/custom models when you want speed, quality, or experiments
- Less ceremony: talk first, decide what to do with it second
- Built on VoiceInk: keeps the strong macOS base, shortcuts, dictionary, app-aware modes, and transcription pipeline
Current status: visible app branding uses roma-just-talk, while many internal labels still come from VoiceInk. The logo and source app icon use the roma-just-talk mark while the fork direction keeps moving first.
- 🎙️ Pre-Roll Capture: Keep a short mic buffer so the start of your thought is not lost
- 🧪 Model Playground: Run local models, cloud providers, or custom endpoints depending on the workflow
- ⚡ Power Mode: Intelligent app detection automatically applies your perfect pre-configured settings based on the app/ URL you're on
- 🧠 Context Aware: Smart AI that understands your screen content and adapts to the context
- 🎯 Global Shortcuts: Configurable keyboard shortcuts for quick recording and push-to-talk functionality
- 📝 Personal Dictionary: Train the AI to understand your unique terminology with custom words, industry terms, and smart text replacements
- 🔄 Smart Modes: Instantly switch between AI-powered modes optimized for different writing styles and contexts
- 🤖 AI Assistant: Built-in voice assistant mode for a quick chatGPT like conversational assistant
Download the latest fork release from GitHub Releases.
The source app icon now uses the roma-just-talk split-keyboard mark. Published downloads may still show older assets until the next release.
Install the fork release via Homebrew:
brew install --cask happyf-weallareeuropean/roma-just-talk/roma-just-talkThe upstream voiceink cask installs upstream VoiceInk, not this fork.
You can build the app yourself by following BUILDING.md.
- macOS 14.4 or later
- Building from Source - Detailed instructions for building the project
- Contributing Guidelines - Original upstream contribution notes
- Code of Conduct - Our community standards
This fork is early. Issues, experiments, and focused patches are welcome when they help the pre-roll dictation direction.
Useful contributions right now:
- Reporting bugs via issues
- Testing pre-roll capture in real macOS writing workflows
- Improving rough docs left over from the upstream project
- Proposing focused changes that make speak-before-hotkey dictation faster, calmer, or more reliable
For build instructions, see BUILDING.md.
This project is licensed under the GNU General Public License v3.0 - learn more.
If you encounter any issues or have questions, please:
- Check the existing issues in the GitHub repository
- Create a new issue if your problem isn't already reported
- Provide as much detail as possible about your environment and the problem
roma-just-talk is built top on: VoiceInk.
- whisper.cpp - High-performance inference of OpenAI's Whisper model
- FluidAudio - Used for Parakeet model implementation
- Sparkle - Keeping VoiceInk up to date
- KeyboardShortcuts - User-customizable keyboard shortcuts
- LaunchAtLogin - Launch at login functionality
- MediaRemoteAdapter - Media playback control during recording
- Zip - File compression and decompression utilities
- SelectedTextKit - A modern macOS library for getting selected text
- Swift Atomics - Low-level atomic operations for thread-safe concurrent programming
- PermissionFlow - make granting permissions intuitive. they were inspired by when codex app shiped legendary
computer-usetool alongside a never seen before way of permission granting flow.
Codex with GPT-5.6 Sol was the primary development agent used throughout Roma Just Talk. It helped inspect and modify the Swift/macOS codebase; implement and debug pre-roll audio and its buffered-to-live streaming handoff; diagnose shortcut, permission, focus, and text-insertion paths; maintain CI and remote end-to-end checks; review and refactor patches; and write documentation and release material.
I remained responsible for the product concept, interaction design, architecture decisions, testing, and deciding which generated changes to accept.
Built from VoiceInk, then pointed at speak-before-hotkey dictation.