shogun-speech-2-text — What's That AI

# shogun-speech-2-text **Built by a shogun who gave up on Windows+H.** Real-time speech-to-text powered by Deepgram Nova-3. Speak into your mic, text appears in any app. That simple. [![GitHub Stars](https://img.shields.io/github/stars/yohey-w/shogun-speech-2-text?style=social)](https://github.com/yohey-w/shogun-speech-2-text) [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT) [![v1.0.1](https://img.shields.io/badge/v1.0.1-Patch_Release-ff6600?style=flat-square)](https://github.com/yohey-w/shogun-speech-2-text/releases/tag/v1.0.1) [![Python](https://img.shields.io/badge/Python-3.10+-blue?logo=python&logoColor=white)](https://python.org) [English](README.md) | [日本語](README_ja.md)

Floating window showing real-time speech recognition

Ctrl+Space → speak → text appears in your active window. $200 free credit included.

Quick Start

Requirements: Python 3.10+, Windows, Deepgram API key (free)

git clone https://github.com/yohey-w/shogun-speech-2-text.git
cd shogun-speech-2-text
install.bat          # Creates venv + installs dependencies
# Edit .env → set DEEPGRAM_API_KEY
start.bat            # Launch!

Press Ctrl+Space, speak, done.

First time? Get your free API key below — takes 30 seconds.

Why This Exists

	Windows+H	shogun-speech-2-text
Accuracy	Mediocre	Excellent
Latency	2-3 sec delay	Near real-time
Tech terms	Butchered	Handled well
Stability	Random freezes	Auto-reconnect
Cost	Free	$200 free (effectively ∞)
Interim results	None	Live display while speaking

Features

Deepgram Nova-3 — State-of-the-art Japanese STT, far superior to Windows+H
Real-time — Interim results displayed as you speak
Clipboard paste — Finalized text auto-pasted via Ctrl+V (IME-safe)
Floating UI — Ctrl+Space toggles a compact always-on-top overlay
Balance display — Remaining Deepgram credits shown in the window
Auto-reconnect — Recovers from WebSocket disconnections automatically
Watchdog — Detects silent connection death (30s timeout)
Custom dictionary (Keyterm) — Register technical terms for better accuracy (configured in .env)

Get Your API Key (Free)

Go to Deepgram Console
Sign up with Google ($200 free credit, no card required)
Create an API key — select Admin role for balance display
Copy the Secret into .env
Ignore the Identifier UUID. It is not the API key, and the Secret is shown only once when you create it.

Select Admin role when creating API key

Select "Admin" under "4. Change role" to enable balance display.

How many Google accounts you have is between you and Google.

Usage

Floating Window (Recommended)

python floating_window.py

Key	Action
Ctrl+Space	Toggle recognition ON/OFF
Esc	Stop & hide window
Ctrl+C	Quit
Drag	Move window

Tray Icon

python tray.py

Console

python main.py

Custom Dictionary (Keyterm Prompting)

Add comma-separated terms to .env:

DEEPGRAM_KEYTERMS=deploy,WebSocket,API,GitHub,Claude

Limit: 500 tokens per request (20-50 terms recommended)
Best for: proper nouns, company names, industry jargon
Leave empty to disable
If you add non-ASCII terms, keep .env saved as UTF-8

Notes

Windows startup registration is being tracked in Issue #1. It is intentionally out of scope for v1.0.1.

File Structure

shogun-speech-2-text/
├── floating_window.py  # Floating UI (recommended)
├── tray.py             # System tray version
├── main.py             # STT core + balance API
├── requirements.txt
├── .env.example
└── docs/
    └── floating_window.png

Cost

Item	Value
Deepgram Nova-3	$0.0059/min ($0.35/h)
Free credit	$200 per Google account
Hours included	~571 hours
At 3h/day	~190 days free

Troubleshooting

pyaudio install fails

Windows:

pip install pipwin && pipwin install pyaudio

Linux/WSL2:

sudo apt-get install portaudio19-dev && pip install pyaudio

Hotkey conflict with IME

Change HOTKEY in floating_window.py:

HOTKEY = "<ctrl>+<alt>+<space>"  # or any combo you prefer

WSL2

Clipboard paste requires running Python on Windows directly. Console-only mode works on WSL2.

License

MIT

</div>

Specs

README