In this paper, I conclude my research findings on audio in machine learning, especially on the speaker recognition system. At first, I briefly givean introduction to the audio domain in the machine learning world and the related attacks. I also present selected papers that I have read in different sub-domains of audio. Then I focus on one of the state-of-the-art speaker recognition systems. I did several experiments related to attacks in the speaker recognition system to test the robustness and vulnerability of an existing system. At last, I present my own attacking system for the speaker recognition system and give some future improvements related to audio in machine learning.