Scanner Transcribe

Real-time browser-based transcription for any audio source. Works with police scanners, radio streams, podcasts, or any browser tab playing audio.

Features

• Dual Transcription Engines
  • Whisper AI - Free, runs locally in browser (~466MB model)
  • Deepgram - Optimized for radio/phone audio ($200 free credit)
• Any Audio Source - Works with any Broadcastify feed or browser tab
• Keyword Alerts - Highlight and get audio alerts on specific words
• Confidence Scoring - Visual indicator of transcription reliability (Deepgram)
• Numeric Normalization - Auto-fixes common number transcription errors
• Tab Audio Capture - Capture audio directly from any browser tab
• Audio Visualizer - Real-time frequency visualization
• Usage Tracking - Monitor transcription time and estimated cost (Deepgram)
• Multi-Format Export - Download transcripts as TXT, JSON, or CSV
• Session Storage - IndexedDB stores all transcripts locally for search and export
• Settings Persistence - Saves feed ID, API key, and keywords locally

How to Use

1. Audio Source

• Enter a Broadcastify feed ID and click "Open Feed", or open any audio source in another tab
• Click "Capture Tab Audio" and select the tab (check "Share tab audio")

2. Transcription Setup

• Choose your engine:
  • Whisper: Click "Load Model" (one-time ~466MB download)
  • Deepgram: Enter your API key (get $200 free credit)
• (Optional) Enter keywords to highlight in transcripts

3. Live Transcript

• Click "Start" to begin transcription
• Download transcripts as TXT, JSON, or CSV

Live Demo

Visit: https://arandomguyhere.github.io/ScannerTranscribe/

Finding Feeds

• Broadcastify: Browse feeds at https://www.broadcastify.com/listen/
• Any Audio: Works with podcasts, YouTube, Twitch, or any tab playing audio

Keyword Alerts

Enter comma-separated keywords to highlight important entries:

• Example: shooting, fire, robbery, ambulance
• Matching entries are highlighted in red
• Audio beep plays when keyword detected

Requirements

• Browser: Chrome or Edge on desktop (required for tab audio capture)
• Mobile: Not supported - mobile browsers cannot capture tab audio
• Internet: Required for audio streams and Deepgram
• Storage: ~466MB for Whisper model (cached in browser)

Transcription Engines

Engine	Accuracy	Cost	Processing
Whisper AI	Good	Free	Local (browser)
Deepgram	Excellent	~$0.0043/min	Cloud API

Deepgram uses the nova-2-phonecall model optimized for radio/phone audio, with keyword boosting for 10-codes and phonetic alphabet.

Export Formats

• TXT - Human-readable with timestamps and keywords
• JSON - Structured data including metadata and keyword list
• CSV - Spreadsheet-compatible format

Technical Details

• Audio Capture: getDisplayMedia() API for tab audio (desktop browsers only)
• Whisper: whisper-small.en model via Transformers.js with police/scanner vocabulary prompts
• Deepgram: WebSocket streaming with nova-2-phonecall model, keyword boosting for 10-codes and phonetics
• Numeric Normalization: Post-processing fixes spaced digits, +1 hallucinations, 10-codes, license plates, and CAD numbers
• Database: IndexedDB for session and transcript storage with full-text search
• Storage: localStorage for settings persistence
• Mobile Detection: Automatic warning for mobile browsers (tab audio capture not supported)

Privacy & Redaction

Enable Redact PII to automatically hide sensitive information:

• [NAME] - Names following "named", "subject", "driver", etc.
• [DOB] - Dates of birth
• [SSN] - Social security numbers
• [PHONE] - Phone numbers
• [PLATE] - License plates
• [ADDRESS] - Street addresses

Enable Normalize Phonetics to convert radio alphabet:

• "Adam Boy Charles 123" → "ABC123"
• Supports LAPD, NATO, and common police phonetics

Accuracy Notes

Transcription accuracy depends on audio quality. Common challenges with scanner audio:

• Numeric data (CAD numbers, plates, addresses) may need human verification
• Low confidence entries (Deepgram) are highlighted in yellow - review these carefully
• Confidence scores below 80% indicate unreliable transcription
• Best results with clear audio and minimal static/crosstalk

Privacy

• Whisper: All processing happens locally in your browser
• Deepgram: Audio is sent to Deepgram's servers for processing
• No tracking: No analytics or third-party tracking

Disclaimer

For informational purposes only. Not for emergency use.