Is OpenDictate a free open-source alternative to Wispr Flow?

Yes. OpenDictate is a 100% free and open-source alternative to proprietary cloud dictation tools like Wispr Flow and Superwhisper. It runs entirely on your local machine with zero subscription fees, zero cloud telemetry, and zero latency.

Does OpenDictate send my voice audio to the cloud?

No. OpenDictate is completely local-first and air-gapped. All neural models (FastConformer, Whisper, Silero VAD) run locally inside ONNX Runtime on your CPU or GPU. Zero audio bytes or telemetry leave your machine.

How does OpenDictate compare to Wispr Flow and Superwhisper in speed?

OpenDictate achieves an ultra-low latency of 80ms using streaming FastConformer CTC models, typing tokens into your active window live as you speak without network roundtrip delays.

Which operating systems are supported by OpenDictate?

OpenDictate supports Linux (via AppImage and deb with /dev/uinput virtual keyboard injection on both Wayland and X11), macOS (Apple Silicon M-series and Intel DMG), and Windows 10/11.

Back to Home

Documentation • User GuideOpenDictate v0.2

OpenDictate User Guide

OpenDictate is a free, open-source, local-first AI voice dictation tool. It converts your speech to text directly on your device with zero cloud latency, zero telemetry, and zero subscription fees.

1. Quickstart (0 to Voice in 60 Seconds)

Step 1: Installation

Linux: Download the .AppImage (or .deb), make it executable (chmod +x OpenDictate*.AppImage), and run it. (Requires /dev/uinput access for global key injection).
macOS: Open the .dmg, drag OpenDictate to Applications. Grant Accessibility and Microphone permissions when prompted.
Windows: Run the installer (.msi or .exe). OpenDictate will start right in your system tray.

Step 2: Initial Setup

On first launch, select your preferred input microphone.
The default Parakeet TDT 110M model (~100 MB) will download automatically in the background.
Once the status indicator shows Ready, you're all set!

Step 3: Your First Dictation

Focus your cursor in any application (VS Code, Notion, Obsidian, Slack, Terminal, Chrome, Word).
Press the global shortcut:
Linux / Windows:Ctrl + Alt + Space
macOS:⌘ + ⇧ + Space
Speak naturally into your microphone.
Press the shortcut again (or pause) — your speech will be instantly typed right at your cursor!

2. Dictation Modes

OpenDictate offers three flexible recording workflows configurable in Settings:

Mode	How it Works	Best For
Push-to-Toggle (Default)	Press shortcut to start recording, speak, and press shortcut again to transcribe & inject.	Writing paragraphs, emails, and long thought dumps.
Hold-to-Talk	Hold down the hotkey while speaking; release the key to immediately inject text.	Quick commands, short chat replies, and rapid coding.
Hands-Free (Wake Word)	Say `“Hey Dictate”` to trigger recording automatically without touching your keyboard.	Cooking, multitasking, or accessibility workflows.

3. Voice Commands & Coding Reference

OpenDictate includes a built-in deterministic command engine for hands-free formatting and text manipulation:

Formatting & Casing

Voice Command	Result	Example Input ➔ Output
"all caps <phrase>"	UPPERCASE	"all caps important notice" ➔ IMPORTANT NOTICE
"camel case <phrase>"	camelCase	"camel case user auth token" ➔ userAuthToken
"snake case <phrase>"	snake_case	"snake case database pool size" ➔ database_pool_size
"kebab case <phrase>"	kebab-case	"kebab case header nav item" ➔ header-nav-item
"title case <phrase>"	Title Case	"title case release announcement" ➔ Release Announcement

Editing & Correction

Voice Command	Action
"scratch that" or "undo"	Reverts the last injected dictation block.
"delete word"	Deletes the previous word before cursor (`Ctrl+Backspace`).
"delete line"	Clears the entire current line.
"clear all"	Clears current input buffer.

Whitespace & Punctuation

Voice Command	Key Output
"new line"	Inserts a single carriage return (`\n`).
"new paragraph"	Inserts two new lines (`\n\n`).
"tab"	Inserts an indentation tab (`\t`).
"bullet point" or "bullet"	Inserts `•` or `-` bullet item.

Smart Workflow Actions

Voice Command	Action
"prompt and send <query>"	Injects query into AI chatbot/terminal and presses `Enter`.
"submit"	Types `Enter` to send form or chat message.
"switch to <application>"	Activates and switches window focus to named application.
"open <url>"	Opens specified web address in your default browser.

4. Custom Vocabulary (Hotword Boosting)

Never have your speech recognizer misspell specialized technical terms or personal names:

Open the Dictionary tab in the OpenDictate dashboard.
Add your custom keywords (e.g., OpenDictate, Kubernetes, FastConformer, GraphQL, Pytorch).
OpenDictate mathematically boosts the acoustic weight for these words during local beam-search decoding, boosting recognition accuracy.

5. Snippet Expansions

Automate repetitive boilerplate text with voice triggers:

Go to the Snippets tab.
Define a Trigger Phrase (e.g., signature, meeting notes, bug template).
Enter the multi-line text template.
During dictation, say “insert snippet signature” — your template expands instantly into the active window!

6. Choosing the Right Speech Model

Full Models Catalog

OpenDictate supports multiple state-of-the-art neural ASR architectures via Sherpa-ONNX:

Model	Disk Size	Latency	Recommended Use
FastConformer CTC (Streaming)	~110 MB	80ms	Fastest. Word-by-word streaming typing as you speak.
Parakeet TDT 110M (int8)	~104 MB	~120ms	Default. Balanced accuracy and speed on any CPU.
Parakeet TDT 0.6B v3	~487 MB	~200ms	High Accuracy. Best for complex terminology and varied accents.
Whisper Turbo (Large v3)	~564 MB	~350ms	Multilingual. Maximum punctuation precision for dictating long articles.

*Switch models at any time with 1 click in the Models tab.

7. Optional AI Voice Polish

If you tend to say “um”, “like”, or ramble, you can enable optional AI polish:

Off (Default)

100% exact verbatim speech transcription with zero alterations.

Clean Mode

Removes verbal filler words, hesitations, and false starts while preserving your exact phrasing.

Bullet Mode

Summarizes raw stream-of-consciousness dictation into clean, formatted Markdown bullet points.

Flexible Engines

Choose between Local SLMs (running completely offline) or Groq Cloud API (for ultra-fast sub-second LLM inference).

8. Privacy & Architecture

🔒 Zero Cloud

Audio processing happens on your local CPU/GPU via ONNX Runtime. Your voice data never leaves your computer.

🚫 No Telemetry

No tracking beacons, session replays, or analytics exist in the codebase.

💾 Local SQLite Storage

Your custom dictionary, snippets, and typing history are stored locally in your OS application directory:

Linux: ~/.local/share/opendictate
macOS: ~/Library/Application Support/opendictate
Windows: %APPDATA%\opendictate

9. Troubleshooting & FAQs

Q: Linux: Hotkey triggers, but no text is typed into the focused window.

A: Ensure your user account has permissions to write to /dev/uinput:

sudo usermod -aG input $USER
echo 'KERNEL=="uinput", MODE="0660", GROUP="input", OPTIONS+="static_node=uinput"' | sudo tee /etc/udev/rules.d/99-uinput.rules
sudo udevadm control --reload-rules && sudo udevadm trigger

Ready to start dictating?

Download OpenDictate v0.2 for Linux, macOS, or Windows and try your first voice dictation in under a minute.

Download OpenDictate Explore Models