How aadityagupta01/ai-voice-detection works
AI Voice Detection is a five-language audio classifier, exposed through an API and accompanied by notebooks and saved artifacts for feature extraction and model training. The primary user workflow is an API request with a Base64 MP3: the service checks access and inputs, decodes and preprocesses audio, predicts a class, and returns confidence and an explanation. The README documents offline audio analysis, MFCC feature extraction, and classical-model training; the tree includes the notebooks and feature arrays. Those notebook internals were not sampled, so their links reflect documented stages only.
Main entry points: main.py Detection Endpoint · ml_feature_extraction.ipynb ML Features · ml_training.ipynb Model Training · ML ML Feature Arrays
Components 16
Offline Modeling
- Audio Analysis
notebooks/audio_analysis.ipynb - ML Features
notebooks/ml_feature_extraction.ipynb - DL Features
notebooks/dl_feature_extraction.ipynb - Model Training
notebooks/ml_training.ipynb
Detection API
- Detection Endpoint
main.py - API Key Check
main.py - Request Validation
main.py - Base64 Decode
main.py - Audio Preprocessing
main.py - Feature Extraction
main.py - Model Inference
main.py - Result Explanation
main.py
Model Assets
- ML Feature Arrays
artifacts/features/ML - DL Feature Arrays
artifacts/features/DL - Trained Model
artifacts/models/best_ml_model.joblib
Other components
- API Caller
Connections 14
- Audio Analysis → ML Features: analyzes audio
- ML Features → ML Feature Arrays: saves features
- DL Features → DL Feature Arrays: saves features
- ML Feature Arrays → Model Training: trains from
- Model Training → Trained Model: selects model
- API Caller → Detection Endpoint: submits request
- Detection Endpoint → API Key Check: checks key
- Detection Endpoint → Request Validation: validates fields
- Detection Endpoint → Base64 Decode: decodes audio
- Detection Endpoint → Audio Preprocessing: processes audio
- Audio Preprocessing → Feature Extraction: extracts features
- Detection Endpoint → Model Inference: predicts class
- Trained Model → Model Inference: loaded at startup
- Detection Endpoint → Result Explanation: formats result