In order to detect the wear and breakage of cutting tools during machining, an extreme learning machine (ELM) model detection method based on sound recognition was proposed. The time-frequency domain characteristics of cutting sound signal were discussed, and the extraction method of cutting tool status-sensitive spectrum energy statistical feature quantity based on wavelet packet decomposition was discussed. A fast ELM detection model based on sound feature quantity recognition was constructed. An example was taken for the identification of cutting wear sound signal in an operation site. The measured data verify that the proposed model can obtain higher detection accuracy and faster response speed. The experimental simulation results show that the ELM model is effective in detecting cutting tool wear with sound recognition.