Hybrid Approach for Accurate Vocal Anomaly Detection Using State-of-the-Art

Accurate voice anomaly detection is achieved by the suggested method's combination of signal processing for feature extraction and improvement with deep learning. Preprocessing, feature extraction, and anomaly detection are the three pillars upon which this approach rests. The raw vocal signal is improved during preprocessing by denoising and filtering. From the preprocessed signal, the feature extraction stage derives Mel-Frequency Cepstral Coefficients (MFCCs) and their derivatives. Finally, an ensemble of three algorithms, including Support Vector Machines (SVM), Long Short-Term Memory (LSTM) Networks, and Convolutional Neural Networks (CNN), is used in the anomaly identification stage to conduct an in-depth evaluation of the detection of voice anomalies. The suggested hybrid method is meant to be more flexible and efficient in practical settings.