Ripple Voice

English · Russian

Voice dictation for Windows. Press — speak — text inserted. Your voice, in flow.

Why

Windows has built-in voice dictation (Win+H), but it:

• Sends your voice to the Microsoft cloud
• Slow and inaccurate
• Doesn't work offline
• Clunky interface

Ripple Voice does the same thing, but:

• Fully offline — the model runs on your computer
• Fast — CUDA GPU accelerates recognition 10x+
• Private — your voice never leaves your computer
• Beautiful — animated HUD shows what's happening

How it works

1. Press the hotkey — HUD appears, recording starts
2. Speak — bars pulse to your voice, the circle shows volume
3. Release the key — recording stops, speech is transcribed
4. Text is inserted into the active field (Notepad, Word, browser — anywhere)

That's it. No windows, no buttons — just speak and text appears.

Installation

git clone https://github.com/Ap3x0s/Ripple-Voice.git
cd Ripple-Voice
python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt

Running

# Via venv (recommended)
Ripple Voice.bat

# Or manually
.venv\Scripts\python.exe main.py

On first launch, the model is downloaded automatically (~1.5 GB). After that it works offline.

GPU (CUDA)

If you have an NVIDIA GPU, the app will automatically use CUDA for faster recognition:

Device	Speed (medium)	Quality
CPU (int8)	~8-12 sec	Excellent
CUDA (float16)	~0.5-1 sec	Excellent

Device selection: Settings → Device → CUDA (GPU) / CPU / Auto

CUDA requirements:

• NVIDIA GPU with CUDA support (GTX 10xx or newer)
• NVIDIA drivers installed
• PyTorch with CUDA (included in installation)

If you encounter cublas64_12.dll not found error, additionally install:

pip install nvidia-cublas-cu12

Usage

Action	Result
Hotkey (hold)	Start recording
Hotkey (release)	Insert text
Escape	Cancel recording
Right-click tray icon	Settings / Exit

Settings

Right-click the tray icon → Settings. In the settings window you can change:

• Hotkey and recording mode (hold / toggle)
• Speech recognition language
• Device (GPU CUDA / CPU)
• Whisper model (tiny → large-v3)
• HUD theme
• Interface language (English / Русский)
• Punctuation

All changes apply immediately after clicking "Save" — no restart needed. Settings are stored in ~/Documents/ripple-voice/settings.json.

HUD Themes

Theme	Style
google	Soft, minimalistic
google_v2	Premium with spring physics
hybrid_v2	★ Default — bars + timer + volume circle
vercel	Sharp, cyber-style

Whisper Models

Model	Size	Speed (CPU)	Speed (CUDA)	Quality
tiny	75 MB	Fast	Instant	Average
base	150 MB	Moderate	Instant	Good
small	500 MB	Moderate	Instant	Excellent
medium	1.5 GB	Slow	~1 sec	Excellent
large-v3	3 GB	Very slow	~1-2 sec	Maximum

Recommendation: medium + CUDA — best balance of speed and quality.

Project Structure

Ripple-Voice/
├── main.py              # Entry point, ties everything together
├── audio_recorder.py    # Microphone audio recording
├── transcriber.py       # Speech recognition (faster-whisper)
├── text_inserter.py     # Text insertion into active window
├── hotkey_manager.py    # Global hotkey interception
├── settings.py          # Settings management
├── settings_window.py   # Settings window (PyQt6)
├── stats.py             # Session statistics
├── history.py           # Recognition history
├── hud_window.py        # HUD theme manager
├── hud_hybrid_v2.py     # Default HUD theme
├── hud_google.py        # Google theme
├── hud_google_v2.py     # Google v2 theme
├── hud_vercel.py        # Vercel theme
├── tray_icon.py         # System tray icon
├── requirements.txt     # Python dependencies
├── Ripple Voice.bat     # Quick launcher (Windows)
└── assets/              # Icons and logos

Requirements

• Windows 10/11
• Python 3.10+
• Microphone
• NVIDIA GPU (optional, for CUDA acceleration)

Privacy

• Speech is processed locally on your computer
• No data is sent to the internet
• The model is downloaded once on first launch
• History and settings are stored only on your machine: ~/Documents/ripple-voice/

License

MIT