Replace microphone with Whisper AI for transcription

- Integrate Transformers.js with Whisper tiny.en model
- Model runs locally in browser (~40MB download, cached)
- Process captured tab audio in 5-second chunks
- Add model loading UI with progress bar
- Filter noise/silence from transcription output
- No microphone needed - transcribes directly from audio stream
- Update README with new workflow

Author: claude

README.md

@@ -1,47 +1,46 @@
 # Baltimore Scanner
 
-Live Police, Fire & EMS radio scanner for Baltimore City with tab audio capture and real-time speech-to-text transcription.
+Live Police, Fire & EMS radio scanner for Baltimore City with Whisper AI transcription running entirely in your browser.
 
 ## Features
 
 - **Tab Audio Capture** - Capture audio directly from the Broadcastify browser tab
-- **Audio Visualizer** - Real-time frequency visualization of the captured audio
+- **Whisper AI Transcription** - OpenAI's Whisper model runs locally in your browser (no server needed)
+- **Audio Visualizer** - Real-time frequency visualization
 - **Volume Control** - Adjust playback volume
-- **Speech-to-Text** - Live transcription using Web Speech API
 - **Timestamped Transcript** - Each entry shows when it was captured
-- **Responsive Design** - Works on desktop and mobile
+- **Privacy First** - All processing happens locally, no data sent to servers
 
 ## How to Use
 
-1. Open the [Broadcastify player](https://www.broadcastify.com/webPlayer/40593) in a new tab
-2. Start playing the audio in the Broadcastify tab
+1. Click **"Load Model"** to download the Whisper AI model (~40MB, one-time download)
+2. Open the [Broadcastify player](https://www.broadcastify.com/webPlayer/40593) in a new tab and start playing
 3. Return to Baltimore Scanner and click **"Capture Tab Audio"**
 4. Select the Broadcastify tab and check **"Share tab audio"**
-5. Click **"Start Transcription"** to begin speech-to-text
-6. The scanner audio will play through this page with live visualization
+5. Click **"Start Transcription"** to begin AI-powered speech-to-text
 
 ## Live Demo
 
 Visit: https://arandomguyhere.github.io/baltimorescanner/
 
 ## Requirements
 
-- **Browser:** Chrome or Edge (required for tab audio capture and speech recognition)
-- **Permissions:** Screen sharing (for tab audio) and microphone (for transcription)
-- **Internet:** Required for Broadcastify stream
+- **Browser:** Chrome or Edge (required for tab audio capture)
+- **Internet:** Required for Broadcastify stream and first-time model download
+- **Storage:** ~40MB for Whisper model (cached in browser)
+
+## Technical Details
+
+- **Audio Capture:** `getDisplayMedia()` API for tab audio
+- **Transcription:** Whisper tiny.en model via [Transformers.js](https://huggingface.co/docs/transformers.js)
+- **Processing:** 5-second audio chunks processed in real-time
+- **No Backend:** Everything runs client-side in WebAssembly
 
 ## Files
 
 - `index.html` - Main page structure
 - `styles.css` - Dark theme styling with visualizer
-- `app.js` - Audio capture, visualization, and speech recognition
-
-## Technical Details
-
-- Uses `getDisplayMedia()` API to capture tab audio
-- Web Audio API for processing and visualization
-- Web Speech API for transcription
-- No server required - runs entirely in the browser
+- `app.js` - Audio capture, Whisper integration, and UI logic
 
 ## Disclaimer