Offline Voice-to-Text to Command Prompt Window or Anywhere Built for Fast Talkers
Dictate is an offline speech-to-text app using OpenAI’s Whisper. I built it because I talk fast and needed something I could tune for my speaking speed that works offline. It works in any application, like a text editor, terminal, console – just hold a hotkey, speak, and release.
Why I Made This
Typing is too slow to get out my thoughts, and sometimes I like to use speech-to-text-tools to work quicker lol, but dont want my voice going to the cloud. So I built Dictate with extensive customization options to handle fast speech patterns n other customizable stuff.
Is it perfect? No. It’s a tool I made for myself that works well enough for my needs. I’m sharing it in case it’s useful to others.
What You Get
- Offline processing – Your voice stays on your machine
- Works everywhere – Any app with text input
- Lots of tuning options – 100+ settings to adjust for your voice
- Calibration wizard – Tests different configurations to find what works
- Two modes – Hold-to-talk or toggle on/off
The Customization Difference
Most voice apps give you basic settings. Dictate exposes all the Whisper parameters so you can tune it for fast speech:
- Beam search settings – Beam size 1-10 (higher = better for fast/complex speech)
- Temperature control – Adjust randomness and fallback behavior
- Detection thresholds – Entropy, compression ratio, no-speech probability
- Quality presets – Fast (75MB), Balanced (142MB), Accurate (466MB)
- Language options – Force English for 30% speed boost vs auto-detect
Calibration Wizard: Runs test phrases, measures accuracy and speed, finds optimal settings for your voice. Takes 15 seconds (quick mode) or 2-3 minutes (full optimization).
Other Features
Hotkeys
- Customizable with Ctrl/Alt/Shift/Win/Cmd modifiers
- Mouse button support
- Per-application overrides
Text Insertion
- Paste mode (fast) or character typing (reliable)
- Auto-detects terminals and uses typing instead of paste
- Restores your clipboard after insertion
Registration
- Online or offline activation
- 2-device limit
- Hardware-bound for security
Common Use Cases
- Dictating prompts to ChatGPT/Claude/other LLMs
- Quick emails and messages
- Code comments (verbatim mode)
- Any text input where typing is slower than talking
⚠️ What This Is (and Isn’t)
This is a personal tool I built for myself. It works for my needs – fast dictation with lots of tuning options and offline privacy.
It’s not perfect:
- Accuracy depends on your voice, speed, and environment
- The calibration helps but doesn’t guarantee perfect transcription
- Support is limited – GitHub issues only, best-effort responses
- Updates happen when I need features, not on a schedule
- May have rough edges or compatibility issues on some systems
Why $5? The calibration tools and full Whisper customization took time to build. Free alternatives exist if you just need basic dictation with default settings that aren’t changeable at all or easily.
This is for people who want simple control, works right out of the box, with optional deep control and are okay with a less polished product.
Technical Details
- Platforms: Windows 10/11, macOS 11+
- Models: 75MB / 142MB / 466MB (downloaded on first use)
- Internet: Required once for product registration and whisper model download, then fully offline
- Languages: 99+ via Whisper
- Requirements: 4GB RAM minimum, CPU with AVX2 support
$5 one-time