README.md

metadata

title: AI Voice Detection API
emoji: 🎤
colorFrom: blue
colorTo: purple
sdk: docker
sdk_version: 4.38.0
app_port: 7860
pinned: false
license: mit

AI Voice Detection API

Detects whether a voice sample is AI-generated or human across multiple languages.

Supported Languages

• Tamil
• English
• Hindi
• Malayalam
• Telugu

Features

• FastAPI-based REST API
• Wav2Vec2 embeddings + signal feature extraction
• Pre-trained classifier for AI/Human voice detection
• Base64 MP3 audio input
• API key protected endpoints

API Endpoints

Health Check

GET /health

Voice Detection

POST /api/voice-detection
Headers:
  x-api-key: <your-api-key>

Body:
{
  "language": "English",
  "audioFormat": "mp3",
  "audioBase64": "<base64-encoded-audio>"
}

Response Format

{
  "status": "success",
  "language": "English",
  "classification": "HUMAN" | "AI_GENERATED",
  "confidenceScore": 0.95,
  "explanation": "Natural prosody, breathing patterns..."
}

Environment Variables

• API_KEY: API key for authentication (default: "hackathon-secret")

Model Architecture

• Uses Facebook's Wav2Vec2-base for audio embeddings
• Extracts 160+ signal features (MFCCs, pitch, spectral, chroma, tonnetz, silence, onset, etc.)
• Ensemble classifier (XGBoost + LightGBM) for final prediction
• Feature-aware explanations using actual acoustic measurements

Team

• ML: Gunashree
• Backend: Tanu
• DevOps/QA: Pavithra