Data Mining for Advanced Anomaly Detection of Speech Using One Class SVM (DaMaDS)

This paper presents DaMaDS: Data Mining for Advanced Anomaly Detection of Speech using One-Class SVM, an innovative framework for detecting anomalies in speech data. The system integrates advanced data preprocessing pipelines, robust feature extraction techniques (such as MFCCs, pitch, and energy), and refined anomaly detection algorithms to identify deviations in speech patterns with high accuracy. Key contributions of DaMaDS include hyperparameter tuning, decision boundary optimization, and multi-speaker normalization, which enhance its performance across diverse datasets. Designed for real-time adaptability, the framework also integrates privacy-preserving mechanisms, making it well-suited for sensitive applications, including speech disorder diagnosis and emergency communication systems. Experimental results demonstrate significant improvements in precision, recall, and F1-score, surpassing traditional methods. By advancing technical innovation and human-centered design, DaMaDS aims to improve communication accessibility and reliability in real-world scenarios, contributing to more inclusive and dependable human-computer interactions.