SpeakPaste

Talk → Text → Paste. Anywhere.

Hold a hotkey, speak, release — your words appear instantly wherever your cursor is.

Download

Grab the latest SpeakPaste.exe — single file, no install.

---

Engines

Engine	Output	Free	Requires
google	Transcribed text	Yes	Nothing
google-cloud	Transcribed text	Free tier	API key
groq	Transcribed text	~8h/day free	API key
google-ext	Transcribed text	Yes	Chrome in background
gemini-lite	English programming prompt	Free tier	Gemini API key
gemini-flash	English programming prompt	Free tier	Gemini API key

Default: google — no key, no setup.

---

Quick Start

Option A — Exe (recommended)

1. Download SpeakPaste.exe from Releases
2. Run it — green icon appears in system tray
3. Right-click → Settings to pick your engine and configure
4. Hold Win+Alt, speak, release — text appears at cursor

Option B — Run from source

git clone https://github.com/mohammad-rj/speakpaste.git
cd speakpaste
python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
python speakpaste.py

---

Settings

All configuration is done via the built-in Settings window (tray → Settings):

• Engine — pick your STT backend; API key field expands inline when needed
• Prompt — off (raw transcript), gemini-lite (transcript → prompt), or gemini-flash (voice → prompt directly)
  • Thinking level — Minimal / Low / Medium / High (default: Low)
  • Media resolution — Low / Medium / High (default: Low)
• Hotkey — default win+alt, change to anything
• Language — e.g. fa, en, ar (or full BCP-47 like fa-IR)
• Follow Windows keyboard layout — when checked, language is detected automatically from your active keyboard layout at the moment you press the hotkey; no manual switching needed (see below)
• Microphone mode — Always-on or On-demand (toggle live from tray)
• Check for updates — notified via tray tooltip on startup

Settings are saved to settings.json next to the exe.

---

Auto language detection

Enable Follow Windows keyboard layout in Settings to let SpeakPaste detect your language automatically.

• Switch to Persian layout with Alt+Shift → hold hotkey → speak Persian
• Switch to English layout → hold hotkey → speak English
• No need to open Settings to change the language — just toggle your keyboard layout as usual

The language is read once when you press the hotkey and stays fixed for the entire recording session. If your layout isn't recognised, it falls back to the language set in the Language field.

Supported layouts: Persian/Farsi, English, Arabic, Turkish, German, French, Russian, Portuguese, Spanish, Japanese, Korean, Chinese.

---

History

Right-click the tray icon and select History to see all transcriptions from the current session.

• Each entry shows the timestamp, engine used, and the output text
• For gemini-lite (two-step processing): both the raw voice transcription and the converted English prompt are shown as separate rows
• Show voice text checkbox toggles the raw STT row for gemini-lite entries
• The list updates in real-time — new entries appear instantly without closing the window
• Clear wipes the session history

History is in-memory only and resets when SpeakPaste is restarted.

---

Engine details

google (default — recommended for most users)

• Google's speech API via SpeechRecognition
• Same engine as Android voice typing — excellent Persian/Farsi support
• Unofficial endpoint, no API key, no Chrome required
• Caveat: unofficial, could change without notice

google-cloud

• Official Google Cloud Speech-to-Text REST API
• Higher accuracy and reliability than the unofficial engine
• Free tier: 60 min/month — sufficient for personal use
• Get a key: console.cloud.google.com → Speech-to-Text API → Credentials

groq

• Records audio → sends to Groq Whisper API
• Free API key, ~8 hours/day limit
• Very accurate, 50+ languages

google-ext

• Chrome Manifest V3 extension with Offscreen Document
• webkitSpeechRecognition running fully hidden in background
• Requires Chrome installed and running
• Setup: chrome://extensions → Developer mode → Load unpacked → select extension/

gemini-lite

• Records audio → Google STT (free) → text → Gemini Flash Lite → English programming prompt
• Speak in any language — output is always a clean English prompt for your AI coding assistant
• Get a free key: aistudio.google.com → Get API key
• System prompt is fully customizable in Settings
• Thinking level and media resolution configurable in Settings (default: Low for both — minimum latency)

gemini-flash

• Records audio → sends WAV directly to Gemini Flash (multimodal) → English programming prompt
• Skips the STT step entirely — Gemini understands voice directly
• Same Gemini API key as gemini-lite; configurable system prompt
• Thinking level and media resolution configurable in Settings (default: Low for both — minimum latency)

---

Microphone mode

Mode	Mic	Pre-roll	Privacy
Always-on	Open all the time	500ms buffer — no cut-off	Mic icon always visible
On-demand	Opens only while hotkey held	None	Closed when idle

Toggle live from tray without restarting.

---

Build from source

pip install pyinstaller
pyinstaller speakpaste.spec

Output: dist/SpeakPaste.exe

License

MIT