Speak-to-Questionnaire

Voice-driven survey answering: reads questions aloud, you speak, say next. BYOK LLM. Works on http pages.

🎙 Speak-to-Questionnaire — answer surveys with your voice An open-source browser extension for voice-driven survey answering: it reads questions aloud, you speak your answers, and say "next" to move on — no typing, no clicking. Built for people who have difficulty typing (visually impaired users, the elderly), and anyone who wants to finish surveys faster. KEY FEATURES • Speak to answer — just say an option ("B", "the second one", or the option text) and it's selected • Voice navigation — say "next", "previous", "repeat", or "skip" to move through the survey; say "done" to confirm multi-choice answers • Question read-aloud — questions are read with text-to-speech; the current question is highlighted on the page so reading and answering states are always clear • Essay dictation — speak freely and it's transcribed to text; optionally an LLM polishes it into written prose (preview first, editable, the original is always preserved) • True/false synonyms — say "true", "correct", "false", or "incorrect" for judgment questions SPEECH RECOGNITION Three recognition modes — the browser's built-in speech service, your own transcription service (OpenAI Whisper-style API), or LLM multimodal transcription. The right one is selected automatically per page, with step-by-step fallback. Recordings are converted to WAV and sent only to the service you configured. BYOK — BRING YOUR OWN KEY AI features use the model service you configure yourself (OpenAI / Anthropic compatible: OpenAI, DeepSeek, Moonshot, Ollama, Dots, and more). Your key is stored only in your browser's local storage — the extension never handles or uploads it. Without an AI key, all core features still work using local rule matching. SUPPORTED PLATFORMS WJX (wjx.cn), Google Forms, Jinshuju, Tencent Survey, Meituan Survey, and any other survey page via "Enable on this page". PERMISSIONS • Storage: saves your local settings only (service endpoint, key, voice preferences) • activeTab + scripting: injects the answering panel into the current survey page, only when you click "Enable on this page" • Offscreen document: handles microphone recording inside the extension's secure context (so http survey pages work too) • Optional host permissions: requested per-domain, only when you save your own service endpoint PRIVACY No developer-side server, no analytics, no data collection. Survey content and audio go only to the services you configured yourself. Full privacy policy: https://github.com/Echo-Smith/speak-to-questionnaire/blob/main/PRIVACY.md Open source (MIT): github.com/Echo-Smith/speak-to-questionnaire — issues and PRs welcome.