<![CDATA[Voice command recognition plays a crucial role in enabling intuitive interaction in robotic and embedded control systems. This study proposes a voice command classification system based on Mel-frequency cepstral coefficients (MFCC) and support vector machine (SVM) using the Google speech commands dataset v2. Eight command classes (“down”, “go”, “left”, “no”, “right”, “stop”, “up”, and “yes”) were used. The dataset was divided into 80 % training and 20 % testing sets, with hyperparameter tuning performed using 5-fold cross-validation on the training data. MFCC feature extraction employed 13 static coefficients augmented with delta and delta-delta features, resulting in a 39-dimensional frame-level representation and a 78-dimensional utterance-level feature vector. Experimental results show that the SVM with radial basis function (RBF) kernel achieved optimal performance with parameters C = 100 and γ = 0.01, yielding 96.2 % accuracy, 96.5 % precision, 96.0 % recall, and 96.2 % F1 score. The inclusion of dynamic features improved accuracy by 4.7 % compared to static MFCCs. The system demonstrates a lightweight architecture suitable for low-resource environments; however, experiments were primarily conducted under clean conditions, and robustness evaluation was limited to a single noise level (20 dB SNR). Furthermore, real-time deployment on embedded hardware was not experimentally validated and remains part of future work.]]>
Copyrights © 2026
