Media
5 items
Description
Talk, get clean text. Whisper runs in your browser, audio never leaves your device. Cleanup, AI prompt structuring, 99 languages. Ideation turns your spoken thoughts into clean, ready-to-use text. Speech recognition runs inside your browser with OpenAI's Whisper, so your audio never leaves your device. Talk through an idea, a message or a prompt for an AI assistant, then paste the result anywhere. WHAT YOU CAN DO • Dictate in any language. The spoken language is detected in the first seconds of each recording, or pin one in Settings. 99 languages with the multilingual models. • Get clean text, not a raw transcript. Filler words, stutters and Whisper artifacts are removed and capitalization is fixed, with language-aware rules for English, Spanish, Portuguese, French, German, Chinese and Japanese. • Turn a ramble into a prompt. One click restructures your stream of consciousness into an organized prompt for the AI assistant you use. A free Google Gemini key is included, or bring your own key for Gemini, OpenAI, Anthropic or Groq. • Smart correction (optional, off by default). An AI model fixes words that were clearly misheard from context, without changing anything else. • Paste or insert. Every transcript is copied to the clipboard automatically. Insert types it into the text field you were in. • Keep a history. Browse and re-copy past transcripts. Stored only in your browser. • See captions while you speak, and the detected language. HOW IT WORKS 1. Press Alt+I (⌃I on Mac) or click the icon to open the side panel. 2. Press Alt+R (⌃R on Mac) or click the mic and talk. 3. Press again to stop. The cleaned transcript is on your clipboard. 4. Paste it, insert it into the page, or click Structure as Prompt. WORKS ANYWHERE YOU TYPE The side panel stays open while you switch tabs, so you can dictate into any web page: chats with AI assistants, email, documents, notes and issue trackers. PRIVATE BY DESIGN • Audio is transcribed on your device. No account, no cloud, no subscription. • No analytics, no telemetry. The extension has no server. • The only optional network use: the one-time model download from Hugging Face (40 to 590 MB depending on the model), and the text, never audio, that you choose to send to an AI provider for prompt structuring or smart correction. • Works offline once the model is downloaded. MODELS Choose in Settings. Tiny and Base run on any machine; Small gives the best accuracy and uses WebGPU when your browser exposes it. LOOKING FOR SOMETHING SIMPLER? Ideation Lite is the same local transcription with no history, settings or AI features, and no network permission beyond the model download. Interface in English, Spanish and Simplified Chinese. Open source (MIT): github.com/tdiipo1/Ideation