The goal of the 2022 Physionet/Cinc Challenge is the development of heart murmurs detector in a patient with various phonocardiograms (PCG). As the 2016 Physionet Challenge successfully proved the combination of a segmenter (that extracts the cardiac cycles) and a classifier (that labels them), a deep learning-based detector is developed using the sequence segmenter-classifier. To enable the global diagnostic, a final algorithm gathering all the patient-related cardiac-cycle labels is added, outputting a unique label per patient. The F. Renna et. al 2019 deep model is used as the segmenter, extracting each cardiac cycle from the PCG with state-of-the-art accuracy. For the classifier, different combinations of input features and deep models are explored. The C. Potes et al 2016 model, that is based on four independent 1D-convolutional feature extractors, it is firstly replicated. Then extensions of this model, increasing the number of input features and varying the processing layers, are also tested. Moreover, an anomaly-detection autoencoder is evaluated, based on the combination of a sequence of 1D-convolutional and pooling layers and a sequence of 1D-convolutional and upsampling layers. For the final diagnostic algorithm, a random forest (RF) and a multilayer perceptron (MLP) are both analyzed using statistical metrics extracted from the arbitrary-lengthen set of cardiac-cycle level labels. The best tested model is a segmenter with an input window length of N=64, the C. Potes classifier with no variations and a final diagnostic MLP. It has been evaluated using 10-fold cross-validation over the public challenge data, scoring the following metrics: 73.7±2.4 % in accuracy, 0.536±0.047 in AUROC, 0.376±0.026 in AUPRC, 28.3±0.5 % in f-measure and 2262±244 in challenge score. Unfortunately, this model could not be tested in the hidden data during the unofficial phase. Note the entry under our team’s name (UZ-ULPGC) was just a test entry with no deep learning models.