Abstract: Tool wear monitoring based on audible signals is promising because of the advantages of low cost, easy installation of sensors, and no interference with the machining process. However, the audible signal contains a large amount of environmental noise, resulting in poor monitoring accuracy. To this end, a tool wear monitoring method based on the entropy characteristics of audible signals is proposed. Firstly, seven kinds of entropy features of audible signals in the machining process are extracted. Secondly, shannon entropy, wavelet entropy, permutation entropy and approximate entropy features with the absolute value of Pearson correlation coefficient (PCC) greater than 0.9 are preferred by correlation analysis. Finally, the tool wear monitoring model is constructed by using the long short-term memory (LSTM) network. The experimental results show that the entropy features can resist noise interference, and the root mean square error (RSME) of monitoring accuracy is 0.004 1 mm.