ImpactScore

An AI-powered speech analysis app that tells you how powerfully your words land — and how to make them unforgettable.

Record or upload a speech, and ImpactScore transcribes it in the browser, sends the transcript to Claude, and reveals an animated score with focused, actionable feedback. Warm peach / cream / rose-gold premium design, soft glassmorphism, slow cinematic animations.

---

What it does

1. Record or upload audio of a speech or presentation.
2. Transcribes it live in the browser using the Web Speech API.
3. Analyzes the transcript with Claude (claude-sonnet-4-20250514).
4. Reveals an animated ImpactScore — a large shimmering number in a circular gauge, four warm-toned dimension bars (Focus · Relevance · Accessibility · Shareability), a top strength and weakness in frosted-glass cards, and three improvement tips.

The scoring methodology is intentionally hidden — the UI only ever shows the score, the four dimensions, and plain-language feedback.

---

Run it

It's a single static HTML file — no build step, no dependencies.

# any static server works, e.g.
python3 -m http.server 8000
# then open http://localhost:8000

Or just open index.html directly in a Chromium-based browser.

Demo mode (no API key needed)

Open index.html#demo to preview the results dashboard with sample data — handy for seeing the UI without recording or a key. This is how docs/dashboard.png was captured.

---

Using it for real

You'll need an Anthropic API key. Paste it into the field on the landing screen; it is used only in your browser to make the analysis request and is never stored or sent anywhere else.

⚠️ Heads-up on the API key. Calling the Anthropic API directly from client-side JavaScript requires the anthropic-dangerous-direct-browser-access header (already included) and means the key is present in the page. That's fine for personal/local use. For anything public-facing, put a small backend in front so the key never ships to the browser.

Browser support: live transcription uses the Web Speech API, available in Chrome and Edge. Uploaded files are transcribed by playing them back while recognition listens; direct recording is the most reliable path.

---

How it works

• Transcription — SpeechRecognition (continuous, interim results, auto-restart for long speeches).
• Analysis — a single POST to https://api.anthropic.com/v1/messages with a hidden system prompt; the JSON response drives the dashboard.
• UI — Cormorant Garamond for headings, Inter for body, glassmorphism cards, a grain texture overlay, a gold-shimmer score reveal, and easing tuned to feel unhurried.

Tech

Plain HTML + CSS + JavaScript. No framework, no build, no backend.

License

MIT