System-Wide Dictation for macOS and Windows
One dictation engine, every application. Works in native apps, browsers, terminals, and even virtual machines.
System-wide dictation means SuperSpeech works anywhere you can type: in Microsoft Word, in your browser, in a terminal, in Slack, in your IDE, in a virtual machine. No per-app integration, no extensions, no plugins to maintain. This is the fundamental architectural choice that makes SuperSpeech practical for daily use.
How system-wide dictation works
Rather than being an app you switch to, SuperSpeech runs as a background service with a global hotkey. When you trigger dictation, audio is captured from your default microphone, transcribed locally, and the resulting text is injected at your text cursor via OS input APIs. On Mac: Accessibility APIs. On Windows: SendInput.
- Works in any text input — no app allowlist
- Runs as a lightweight tray/menu bar app (~100MB RAM)
- Global hotkey triggers dictation without app focus change
- Output goes to your current cursor position
- No extension, plugin, or per-app configuration
Tested apps and environments
SuperSpeech works reliably in:
- Microsoft Office (Word, Excel, PowerPoint, Outlook)
- Google Docs, Sheets, Slides, Gmail
- Slack, Discord, Teams, Zoom chat
- VS Code, JetBrains IDEs, Sublime, Zed, Xcode
- Chrome, Safari, Firefox, Edge — all web apps
- Terminal, iTerm2, Warp, Windows Terminal, PowerShell
- Notion, Obsidian, Craft, Bear, Drafts
- Electron apps and VM environments (Parallels, VMware)
Architecture
SuperSpeech is built as three layers: native tray UI (AppKit/SwiftUI on Mac, WinUI/WPF on Windows), a ring-buffered audio engine running at 16kHz mono, and a ML inference engine using Core ML on Mac or ONNX Runtime on Windows. These layers communicate via in-process channels for zero IPC latency.
Resource footprint
Idle: ~100MB RAM, 0-1% CPU. During dictation: adds ~400MB RAM with model loaded, 15-20% CPU burst during inference (often less on GPU/ANE). This makes SuperSpeech comfortable to leave running all day.
Frequently Asked Questions
Does it work in web apps?
Yes. System-wide dictation works in any text input field in any browser — including Google Docs, Notion, Linear, Slack web, and any webapp.
Does it need extensions or plugins?
No. No browser extensions, no app plugins. SuperSpeech works at the OS input level, so it functions anywhere typing works.
Does it work in virtual machines?
Yes, with some caveats. On Parallels Desktop, Accessibility forwarding needs to be enabled. Windows VMs on Mac work via SendInput forwarding.
How much RAM does it use?
Idle: ~100MB. With model loaded and dictating: ~500MB. Model unloads after extended idle to free memory.
Does it slow down my Mac or PC?
Idle impact is negligible. During dictation, brief CPU/GPU burst for inference, then idle again. Most users do not notice SuperSpeech running.
One tool for every text input
Works in Word, Gmail, Slack, VS Code, and everywhere else. Try it free for 30 days.