README.md

metadata

title: Voice Assistant - Multi-language Division Matching & Contact Search
emoji: 🎙️
colorFrom: purple
colorTo: blue
sdk: gradio
sdk_version: 4.0.0
app_file: app.py
pinned: false
license: mit

🎙️ Voice Assistant Demo

A powerful multi-language voice assistant that helps users find divisions and contacts within an organization using natural language queries.

🌟 Features

🗣️ Multi-language Voice Input

• 99+ languages supported (auto-detected)
• Automatic speech-to-text using OpenAI Whisper
• Arabic-to-English translation for seamless processing
• Works with various audio formats

🎯 Smart Division Matching

• Semantic search using sentence embeddings
• Confidence-based routing with intelligent thresholds
• Department-level expansion (searches all divisions in a department)
• Fast matching (~50ms) using all-MiniLM-L6-v2

👤 Name Extraction

• Extracts person names from queries using GLiNER
• Supports English and Arabic names
• Zero-shot NER for robust extraction

📞 Contact Search

• 500+ contacts across 23 departments and 67 divisions
• Intelligent matching combining name and division
• Confidence scoring with match reasoning
• Fuzzy name matching for typos and variations

🚀 How to Use

Division Matching (Text)

Find the right division for your query:

"I need help from IT Security"
"Find someone in Finance"
"Connect me to Human Resources"

Division Matching (Voice)

Speak your query in any language - it will be transcribed and processed automatically.

Contact Search (Text)

Search for specific people or teams:

"Find Dima in Information Technology"
"Ahmed Al-Malek"
"I need to talk to someone in Legal"

Contact Search (Voice)

Speak your contact search query in any language.

📊 Example Queries

Department-Level Queries

These queries search across ALL divisions in a department:

• ✅ "Find someone in Information Technology" → Searches 8 IT divisions
• ✅ "I need help from Finance" → Searches all Finance divisions
• ✅ "Connect me to Human Resources" → Searches all HR divisions

Division-Level Queries

These match specific divisions:

• ✅ "Find Ahmed in App Dev" → Applications Development & Integrations
• ✅ "I need help from IT Security" → IT Security Implementation & Operations
• ✅ "Connect me to Legal" → Legal divisions

Name-Only Queries

• ✅ "Find Dima" → Searches all contacts named Dima
• ✅ "Ahmed Al-Malek" → Exact name match
• ✅ "I need to talk to Rashed" → Fuzzy name matching

Combined Queries (Name + Department/Division)

Priority given to division accuracy:

• ✅ "Find Dima in Information Technology" → Perfect match (confidence: 1.00)
• ✅ "Find Ahmed in App Dev" → Shows App Dev team members

🔧 Technical Details

Models Used

• Embeddings: sentence-transformers/all-MiniLM-L6-v2 - Fast, lightweight semantic search
• Name Extraction: urchade/gliner_small-v2.1 - Zero-shot NER for person names
• Speech-to-Text: openai/whisper-base - Optimized for CPU with good accuracy

Confidence Scoring

Score	Meaning	Example
1.00	Perfect match (name + division)	Dima in IT
0.95	Exact name match	Ahmed Al-Malek
0.66	Strong division match	People in requested division
0.59	Good division match	Close division match
< 0.30	Low confidence	Wrong division penalty

Match Reasons

• name_and_division_match - Both name AND division match ✅
• division_match - Division/department matches (no name match)
• exact_name_match - Exact name match (100%)
• fuzzy_name_match - Partial name match (75%+)
• name_match_wrong_division - Name matches but WRONG division ⚠️

📦 Database Stats

• 500 contacts across the organization
• 23 departments (Information Technology, Finance, HR, etc.)
• 67 divisions (specific teams and units)
• Multi-language support (English + Arabic names)

🌍 Supported Languages

The voice assistant supports 99+ languages including:

• English
• Arabic (العربية)
• Spanish, French, German, Italian
• Chinese (中文), Japanese (日本語), Korean (한국어)
• Hindi, Urdu, Bengali
• And many more...

Language is automatically detected - just speak naturally!

⚡ Performance

• Division Matching: ~50ms per query
• Name Extraction: ~100-200ms per query
• Voice Processing: ~1-3 seconds (depends on audio length)
• Contact Search: ~100-300ms per query

🛠️ Built With

• Gradio - Interactive web interface
• FastAPI - Backend API (original implementation)
• Sentence Transformers - Semantic search
• OpenAI Whisper - Speech recognition
• GLiNER - Named Entity Recognition
• PyTorch - Deep learning framework

📝 License

MIT License

🙏 Acknowledgments

• OpenAI for Whisper
• Hugging Face for model hosting
• URCHADE for GLiNER
• Sentence Transformers team

---

Version: 4.0.0 Status: ✅ Production Ready Demo Type: Interactive Gradio Demo