AI Generated Voice Detection API

This project provides a FastAPI-based backend service that detects whether a voice sample is human or AI-generated using machine learning and audio feature analysis.

Overview

The system accepts base64-encoded MP3 audio, extracts statistical voice features, scales them using a pretrained scaler, and performs classification using a TensorFlow / Keras model.

The API returns:

• Voice classification (HUMAN / AI_GENERATED)
• Confidence score
• Human-readable explanation

Features

• FastAPI REST API
• API key authentication
• Base64 audio input support
• MP3 audio validation
• Audio feature extraction
• Pretrained ML model inference
• Confidence scoring
• Explanation generation based on vocal characteristics

Tech Stack

• Python
• FastAPI
• TensorFlow / Keras
• Librosa (via feature extractor)
• Scikit-learn
• Joblib

API Endpoint

POST /api/voice-detection

Headers

x-api-key : YOUR_API_KEY

Request Body

{
  "language": "english",
  "audioFormat": "mp3",
  "audioBase64": "BASE64_ENCODED_AUDIO"
}

Response Example

{
  "status": "success",
  "language": "English",
  "classification": "AI_GENERATED",
  "confidenceScore": 0.87,
  "explanation": "Speech shows low pitch variation, commonly observed in synthetic voices."
}

Error Response Example

{
  "status": "error",
  "message": "Only mp3 audio is supported"
}

Supported Languages

• English
• Hindi
• Tamil
• Telugu
• Malayalam