Audio Recognition with Graphical Interface-Based Model for Assisted Robotics

This paper presents an optimized audio recognition system designed for a personal database recorded in Romanian by both male and female voices, aiming to assist elderly individuals through integration with assistive robotics. The study explores a structured pipeline, including dataset preparation, feature extraction like Mel-Frequency Cepstrum Coefficients (MFCC)-34, MFCC-38, Spectrogram, and Mel Spectrogram and classification using Support Vector Machine (SVM), k-Nearest Neighbor (k-NN) and Convolutional Neuronal Network (CNN). A majority voting approach is employed, leveraging six models, with CNN using Spectrogram receiving two votes, demonstrating superior performance in classifying test samples and even unfamiliar voices. Performance evaluation is conducted using a confusion matrix and multiple metrics to ensure robustness. To enhance usability, a Graphic User Interface (GUI) implemented in Python enables users to select an audio signal, extract features, classify the input, and display the final decision based on majority voting. Additionally, the classification process is demonstrated in the Python console. Furthermore, this paper presents how the integration with Universal Robot UR5, an assistive robot is established, enabling real-world deployment for elderly care in indoor environments.