Command Words Recognition Algorithm of Shrinking Residual Network based on MFCC and Dual Micro-Array

摘要

In order to improve the accuracy and robustness of command word recognition in complex environments, this paper studies a command word recognition algorithm based on Shrinking residual network combining dual microarray and Mel-cepstrum coefficients. The ResNet15 network is improved by using dual micro-array datasets and contraction residual units. two multi-task contraction residual models RSN-CW with command word recognition and user judgment systems are constructed. Among them, the RSN-CW15 overall command word recognition rate and The accuracy of user judgment both exceeding the ResNet15 model. Compared with ResNet15, the training parameters of low-power RSN-CW6 are greatly reduced while ensuring accuracy.