The paper describes the dataset used for building machine learning models for labeling respiratory rate signals into four classes: breath-in, breath-out, and retentions after inhale and exhale. Additionally, we introduce a label to represent segments of the signal infected by noise. The data was collected simultaneously using different types of sensors: a tensometer and two accelerometers. The datasets have been made publicly available via the Gdansk University of Technology repository "Most Wiedzy", ensuring open access to the data and reproducibility of research on respiratory classification. Along with the data we also publish the source files of tools used for building the datasets as well as our implementation of the models for respiratory rate classification and visualization. The data have been stored in CSV format and organized through a directory structure according to different breath patterns. These datasets can be easily processed and converted for usage with different machine learning methods across various research applications in respiratory health.