README.md

metadata

title: ShiblASR-v1.0
emoji: 🎙️
colorFrom: purple
colorTo: gray
sdk: gradio
sdk_version: 5.49.1
app_file: asr_disfluency.py
pinned: false
license: mit

🎙️ ShiblASR v1.0 — Verbatim Disfluent Transcription

Multilingual ASR system with advanced disfluency detection, pause timestamping, and grammar analysis.

Features

✅ Multilingual Support - All Whisper-supported languages
✅ Disfluency Detection - Fillers, pauses, repetitions
✅ Pause Timestamping - Exact timing of silent periods
✅ Grammar Analysis - Highlighted grammar issues
✅ Offline Processing - No external API calls
✅ Export Results - HTML and TXT downloads

Usage

1. Record or upload audio file (WAV, MP3, etc.)
2. Click "Start Transcription"
3. View results in two tabs:
   • Verbatim: Original transcription with disfluencies highlighted
   • Clean: Grammar-corrected version
4. Download results for further analysis

Output Details

• 🌐 Detected Language: Automatically identified language
• ⏸️ Detected Pauses: Timestamped silence periods (≥3 seconds)
• 🗣️ Verbatim Transcript:
  • [filler] - Detected filler words
  • [pause] - Short pauses (3-5 seconds)
  • [LONG PAUSE] - Extended silences (>5 seconds)
  • grammar - Highlighted grammar issues
• 🧠 Clean Transcript: Grammar-corrected text

Technical Details

• Based on OpenAI Whisper Base model
• PyTorch backend (CPU/GPU compatible)
• LanguageTool for grammar analysis
• LibROSA for audio processing