Bowel Sound Detection Based on MFCC Feature and LSTM Neural Network

A method of detecting the bowel sound (BS) by LSTM neural network using MFCC features has been proposed in this paper. A large amount of bowel sound data has been recorded. The MFCC features were extracted from framed BS data, which were used to train a neural network based on LSTM. The method was tested in different situations. When the recording environment of the test data was the same with that of the training data, this method can achieve the sensitivity of 90.92% and total accuracy of 92.56%. When the recording environment was different, the sensitivity was affected and drops to 62.1%, but the specificity was still kept at 96.2%, and the total accuracy at 94.2%. This first attempt of applying the method of voice recognition to bowel sound detection has been proved to be a success. This method not only can be used to distinguish bowel sound audio frames and noise frames with the state of the art accuracy, but also can detect the time when bowel sound occurred in a piece of audio.