In industrial manufacturing processes, such as plugging small connectors in, where visual verification is difficult, workers may experience difficulties in detecting failures. Artificial intelligence algorithms can be used to detect and identify the sound of these connectors and mitigate human error. In this work, sound samples of correctly plugged-in connectors and ordinary background noise of the workplace were collected using a recording setup fastened to workers' hand. In order to discriminate anomalies that represent failures, autoencoder models were trained and tested in an unsupervised manner. Experiments with different deep learning architectures for anomaly detection are conducted. Our CNNAE-FT model achieved best results and yielded a ROC-AUC score of 0.85.