AI Voice Detection API - HCL GUVI Buildathon

Multi-language AI voice detection system that determines whether a voice sample is AI-generated or human-generated.

Features

• Supports 5 languages: Tamil, English, Hindi, Malayalam, Telugu
• Base64-encoded MP3 input
• JSON response with classification, confidence, and explanation
• API key authentication
• Language-agnostic audio feature analysis
• REST API with FastAPI

API Specification

Endpoint

POST /detect-voice

Headers

X-API-Key: HCL_GUVI_2024_VOICE_DETECTION_KEY
Content-Type: application/json

Request Body

{
  "audio_base64": "BASE64_ENCODED_MP3_AUDIO"
}

Response

{
  "classification": "AI_GENERATED" or "HUMAN",
  "confidence_score": 0.85,
  "explanation": "Analysis detected: Very consistent spectral characteristics; Low formant variation; Unnaturally smooth pitch contour",
  "language_detected": null,
  "processing_time_ms": 234
}

Detection Method

The system analyzes multiple audio features:

1. MFCC (Mel-frequency cepstral coefficients) - Spectral envelope analysis
2. Spectral Features - Centroid, rolloff, bandwidth consistency
3. Pitch Variation - AI voices have unnaturally smooth pitch
4. Energy Consistency - AI maintains too-consistent energy levels
5. Zero Crossing Rate - Noise pattern analysis
6. Formant Patterns - Natural vs artificial voice characteristics

Key AI Voice Indicators:

• Too consistent spectral characteristics
• Unnaturally smooth pitch contour
• Highly consistent energy levels
• Controlled spectral bandwidth
• Lack of natural noise/breathing patterns

Key Human Voice Indicators:

• Natural spectral variation
• Natural pitch fluctuations
• Energy variation (breathing, pauses)
• Natural noise patterns
• Formant pattern variation

Quick Start - Local Testing

1. Install Dependencies

pip install -r requirements.txt

2. Run the Server

python app.py

The server will start at http://localhost:8000

3. Test the API

# Visit the interactive docs
http://localhost:8000/docs

# Or use the test script
python test_api.py

the /docs endpoint for interactive API documentation.