# LinguaVox — Full Product Reference

> AI-powered voice transcription and translation Chrome extension. Speak into any web input field, any language. Under 3 seconds end-to-end.

## What is LinguaVox

LinguaVox is a Chrome browser extension (Manifest V3) that converts speech to text using OpenAI Whisper and injects the result directly into any web input field — chat boxes, search bars, email editors, CRM forms, support consoles, documentation tools, and more.

The core interaction is designed to be frictionless: the user holds Ctrl+Space, speaks, releases the shortcut, and the transcribed (and optionally enhanced) text appears in the active field within under 3 seconds. No copy-pasting, no switching windows, no separate dictation app.

All OpenAI API calls are proxied through a Cloudflare Worker running at the global edge. API keys never reach the browser. The extension authenticates users via Google OAuth 2.0 and issues a short-lived JWT for each session.

## How Recording Works

The recording flow has two entry points depending on the target website:

**Standard websites (Google, Gmail, most apps):**
1. User presses Ctrl+Space — content script detects the keydown event
2. Extension checks microphone permission, shows a recording indicator (top-right corner, red dot + timer)
3. Background service worker activates the browser's MediaRecorder API
4. User speaks
5. User releases Space — content script hides the indicator, shows a processing spinner
6. Background worker stops recording, collects the audio blob as base64
7. Audio is sent to the Cloudflare Worker endpoint /api/transcribe
8. Cloudflare Worker calls OpenAI Whisper, returns transcribed text
9. If AI enhancement is enabled, text is sent to /api/enhance (GPT-4o-mini)
10. Final text is injected into the active input field via the content script
11. A brief "Text inserted" confirmation toast appears

**Complex SPAs (Slack, Asana — intercept keyboard events):**
The extension also listens via chrome.commands.onCommand. When the keyboard shortcut fires through the commands API (bypassing page-level interception), the background service worker orchestrates the same recording flow and sends a message to the content script to show/hide UI indicators.

The end-to-end latency target is under 3 seconds from releasing the shortcut to text appearing. In practice, this depends on audio length and network conditions, but short utterances (5–15 seconds) consistently hit the target.

## Supported Languages

LinguaVox supports 21+ languages for both transcription (speech-to-text) and translation (speak in one language, text appears in another):

Transcription languages: English, Ukrainian, Russian, Spanish, French, German, Japanese, Korean, Chinese (Simplified), Arabic, Italian, Portuguese, Polish, Dutch, Turkish, Hindi, Swedish, Romanian, Greek, Tagalog, Punjabi

Translation target languages: Same set — users can speak in any supported language and have the output appear in a different language. For example, speak in Ukrainian, get English text in the input field.

The interface (extension popup, notifications, tooltips) is available in 14 locales: English, Ukrainian, Russian, Spanish, French, German, Japanese, Chinese, Romanian, Greek, Urdu, Tagalog, Italian, Punjabi.

## AI Enhancement Modes

After transcription, LinguaVox can optionally pass the raw Whisper output through GPT-4o-mini for text improvement. There are 6 enhancement modes:

- **Smart Polish** — default mode; fixes grammar, improves clarity, removes filler words, preserves the original meaning and tone
- **Business Style** — rewrites text in professional business tone suitable for corporate emails, reports, and formal communications
- **Grammar Fix** — minimal intervention; corrects spelling and grammar errors only, does not change style or vocabulary
- **Creative Style** — rewrites with vivid, engaging language; adds descriptive detail and stylistic flair
- **Casual Style** — friendly, conversational tone; removes formality, makes text feel natural and approachable
- **Academic Style** — formal academic language; precise vocabulary, structured sentences, citation-ready phrasing

Enhancement is opt-in and runs server-side. If the enhancement API call fails for any reason, the original Whisper transcription is preserved and inserted — enhancement failure never causes a total failure.

Users can also enable "Original Mode" which bypasses enhancement entirely and inserts the raw transcription, regardless of which enhancement mode is selected.

## Text Insertion Compatibility

LinguaVox injects text into the active web input using a compatibility stack that supports modern web frameworks:

1. **Tracked active element** — the extension monitors focus and click events to always know which field is active
2. **React-compatible setter** — uses the native input value setter plus a synthetic InputEvent with inputType:'insertText', triggering React's onChange handlers correctly
3. **ContentEditable support** — for rich text editors (Slack message box, Notion, Gmail compose), uses document.execCommand('insertText') with a DOM mutation fallback
4. **Clipboard fallback** — if all else fails, copies text to clipboard and notifies the user

This stack has been tested and confirmed working on: Google Search, Gmail compose, Slack message input, Asana task fields, Notion pages, standard HTML input and textarea elements, Draft.js editors, Quill editors, and most CRM/support tool inputs.

## Privacy and Security

- Audio is never stored. It is processed in real-time and discarded after transcription.
- No personal data is sold or shared with third parties.
- OpenAI API keys are stored exclusively in Cloudflare Worker secrets — the browser never sees them.
- User identity (Google OAuth profile) is stored in Cloudflare KV as: Google ID, email, name, account creation date, and active status.
- Daily usage counts are stored with a 25-hour TTL and automatically expire.
- All communication between the extension and the Worker uses HTTPS with JWT Bearer authentication.
- The extension requests only the permissions it needs: activeTab, storage, scripting, identity, notifications, tabs.

## Pricing and Rate Limits

**Free tier:**
- 20 requests per day (combined transcription + translation + enhancement calls)
- Shared OpenAI API key pool managed by LinguaVox
- No credit card required, sign in with Google

**Bring Your Own Key (BYOK):**
- Unlimited requests, billed directly to the user's OpenAI account
- User provides their OpenAI API key in the extension settings
- Key is sent to the Cloudflare Worker per request and used for that session only — not stored server-side

**Organization plan:**
- Team-level API key sharing
- Usage analytics across team members
- Centralized billing through one OpenAI account
- Admin dashboard for usage monitoring

Rate limit errors return HTTP 429 with a JSON body: `{"error": "Daily limit exceeded", "limit": 20, "used": 20}`. The extension surfaces this as a user-facing notification.

## Technical Architecture

**Chrome Extension (Manifest V3):**
- background.js — service worker; handles recording orchestration, API calls, auth state
- content.js — injected into every page; keyboard handling, recording UI, text insertion
- popup.js / popup.html — settings UI, history viewer, auth flow

**Cloudflare Worker (serverless, global edge):**
- /api/auth/google — Google OAuth token exchange, JWT issuance
- /api/user/me — returns user profile and current daily usage count
- /api/transcribe — forwards audio to OpenAI Whisper, returns text
- /api/enhance — forwards text to GPT-4o-mini with selected enhancement prompt
- /api/translate — translates text to target language via GPT-4o-mini

The Worker rotates through a pool of OpenAI API keys automatically on 429 (rate limit) or 401 (invalid key) errors, ensuring high availability even when individual keys hit limits.

**Infrastructure:**
- Cloudflare Worker for API proxy and auth
- Cloudflare KV for user accounts and usage tracking
- Cloudflare Pages for landing page hosting
- OpenAI Whisper (large-v2 equivalent) for transcription
- OpenAI GPT-4o-mini for enhancement and translation

## Frequently Asked Questions

**Does LinguaVox work on all websites?**
It works on the vast majority of websites including Google, Gmail, Slack, Asana, Notion, Salesforce, Jira, and standard HTML forms. Google Docs and Google Slides have a complex editor implementation that may require a page refresh to activate correctly.

**Do I need an OpenAI account?**
No. The free tier uses a shared API key pool. You only need your own OpenAI key if you want unlimited usage beyond 20 requests/day.

**Is my voice recorded or stored?**
No. Audio is streamed to the Cloudflare Worker, immediately forwarded to OpenAI Whisper, and discarded. There is no audio storage at any point.

**What keyboard shortcut does it use?**
Default is Ctrl+Space. The shortcut can be reconfigured in Chrome's extension keyboard shortcut settings (chrome://extensions/shortcuts).

**How accurate is the transcription?**
OpenAI Whisper achieves 95%+ word error rate accuracy on most languages in typical speaking conditions. Accuracy improves with clear speech, a decent microphone, and minimal background noise.

**Does it work offline?**
No. Transcription requires sending audio to OpenAI Whisper via the Cloudflare Worker. An active internet connection is required.

**What happens if the AI enhancement fails?**
The original Whisper transcription is preserved and inserted into the field. Enhancement failure never causes the transcription to be lost.

**Can I use it in incognito mode?**
Yes, if you enable the extension for incognito mode in Chrome's extension settings. Auth state is stored in chrome.storage.local which does not persist across incognito sessions.

## Contact and Links

- Website: https://linguavox.uk
- User Dashboard: https://linguavox.uk/login
- Privacy Policy: https://linguavox.uk/privacy/
- Support: https://linguavox.uk/support/
- GitHub: https://github.com/kos-4862/linguavox-public
- Chrome Web Store: https://chromewebstore.google.com/detail/linguavox/dpdejiobdhljljfnkmipjblbpcfhbdea

## Version Information

Current version: 3.0.0
Last updated: 2026-05-07
Platform: Chrome Extension (Manifest V3)
Minimum Chrome version: 88+