© 2018 IEEE. This paper is mainly about the accuracy on the children speech recognition under noisy circumstances and methods to improve it. The power of the children's speech is lower than the adult. The accuracy of children speech recognition is unavoidably affected by the ambient noise. We executed a series of experiments to calculate the recognition accuracy for children speech under clean condition and different noisy conditions. It has been reported that running spectrum analysis (RSA) and running spectrum filtering (RSF) have the capability to enhance the robustness in the speech recognition system under noisy conditions. In subsequent experiments, we added RSA and RSF to the speech feature extraction and obtained a children recognition system with enhanced noise suppression capability and achieved higher recognition accuracy.