Randomized trial evaluates voice command recognition in mobile robotic systems, suggesting improved interaction methods.
Key Points
This study aims to develop a voice command classification system to enable intuitive control of mobile robots.
Implemented a classification system using mel-frequency cepstral coefficients (MFCC) and support vector machines (SVM) with the Google speech commands dataset v2.
Divided the dataset into 80% training and 20% testing, performing hyperparameter tuning through 5-fold cross-validation.
Utilized 13 static MFCC coefficients with dynamic features to create 39-dimensional frame-level and 78-dimensional utterance-level vectors.
The SVM with radial basis function kernel achieved 96.2% accuracy, 96.5% precision, 96.0% recall, and 96.2% F1 score with parameters C=100 and γ=0.01.
Incorporating dynamic features improved accuracy by 4.7% over static MFCCs.
Evaluation was primarily under clean conditions, with limited robustness testing at a single noise level of 20 dB SNR.