The Respeck is a wireless sensor device is worn as a plaster on the chest to measure continuously features of pulmonary function such as the respiratory rate and respiratory flow/effort. Given the constraints of the location of the Respeck on the body, this paper describes deep learning methods - firstly, to classify five different types of walking: shuffle walking, normal walking, running, ascending/descending stairs, and, secondly, to count the number of steps (step-count) based on methods using representation of the sensor data in the time and frequency domains to count the number of steps. Results are presented for the accuracy of greater than 90% for the classification of walking-types, and step-counts. The methods presented in this paper compare favourably (within 2%) with commercial wrist-worn pedometers, such as the Apple Watch.