Biometric Security & AudioAIFB 2025 / University Project
Edge-Based Voice Biometric Verification & Cloud Anti-Spoofing
Constructed a speaker recognition system combining edge-based audio acquisition with a centralized cloud backend, achieving 97.5% authentication accuracy in controlled environments.
Authentication Accuracy
97.5%
Architecture
Edge Capture + Cloud Inference
Feature Extraction
MFCCs & Spectral Jitter
Backend Framework
FastAPI + Librosa
Abstract & Overview
Voice biometric authentication systems are vulnerable to replay and synthetic voice cloning attacks. This architecture pairs low-latency edge-based audio capture and MFCC extraction with cloud-based anti-spoofing verification powered by FastAPI and Gaussian Naive Bayes classification, hitting 97.5% authentication precision.
Problem Formulation
Consumer IoT and edge verification hardware have constrained compute budgets, rendering complex multi-billion-parameter neural models impractical for on-device voice authentication.
Architecture & Key Mechanisms
1Edge Audio Pipeline: Captures clean audio streams and extracts 13-dimensional Mel-Frequency Cepstral Coefficients (MFCCs).
2Centralized Anti-Spoofing: Dispatches acoustic signatures to a high-throughput FastAPI backend for probabilistic verification.
3Spectral Variance Defense: Validates natural vocal tract jitter and spectral frequency anomalies against synthetic voice models.
Empirical Highlights
- Formulated embedding pipeline utilizing MFCCs and spectral features via Gaussian Naive Bayes classifier.
- Engineered FastAPI backend and web interface facilitating secure user enrollment and parallelized inference.
- Evaluated against environmental noise and spoofing vulnerabilities.
Technologies & Tooling
PythonFastAPILibrosaMFCCsOpenCVScikit-learn