Speech Command Recognition via Ensemble Learning of MS-LFB and MFCC Features

Le résumé fourni par la source

Voice-controlled systems need speech command recognition features to provide humans with a seamless interaction experience. Research examines the performance of individual Mel-Scaled Log Filter Bank (MS-LFB) and combined MS-LFB and Mel Frequency Cepstral Coefficients (MFCC) features within speech command detection systems. Using the TensorFlow Speech Recognition Challenge dataset containing target commands and silence along with unknown inputs this investigation examines data preprocessing processes and feature extraction approaches combined with noise enhancement techniques to strengthen model capabilities within practical environments. A convolutional neural network architecture processes MS-LFB and MFCC features independently before an ensemble system of the individual outputs generates improved classification accuracy results. The results show that the MS-LFB model achieved a validation accuracy of 91.51%, outperforming the MFCC model by 3.2%. However, the combined ensemble model delivered the best performance, reaching a validation accuracy of 93.01%. This proves that the two feature sets work cohesively to enhance classification accuracy, especially in imbalanced datasets. The findings demonstrate that combining different audio feature representations through ensemble learning is a highly effective method for building reliable and robust speech recognition systems.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.