Convolutional Neural Network-based Keyword Classification for Mixer Control

This study proposed a keyword recognition of microphone feed-in, achieving the effect of allowing users to control the real-time mixer system intelligently. The proposed system structure includes a front-end circuit to capture voice and Mel-scale Frequency Cepstral Coefficients (MFCC) features in conjunction with a Convolutional Neural Network (CNN) algorithm. The testing results show that the lightweight CNN model achieves a keyword recognition accuracy rate of 96.4%.