Chronic respiratory diseases, such as asthma and chronic obstructive pulmonary disease (COPD) are significant and growing causes of morbidity and mortality worldwide. Poor disease control causes a substantial burden on patients, their families and society. Inhaler therapy is the most popular asthma and COPD treatment, but patients are compromised from inappropriate medication use. How to measure adherence to medication is a huge global problem. This thesis presents an original audio-based monitoring system for the pressurised metered-dose inhaler (pMDI), which has been developed to track and analyse the patient's dose delivery and inhaler technique. Prior to this study, there existed no automated system, based on acoustic monitoring, to track and assess pMDI inhaler technique. In terms of the hardware, this system combines accelerometer and acoustic sensors to enable a comprehensive assessment of the pMDI technique. For recognition of the breath phases, this research initially employed a hidden Markov model (HMM) with a Gaussian mixture model (GMM) to identify the phases of the breath sound. Ultimately a model was developed that concatenating two deep learning models (1-D ResNet18 and CLDNNs), which improved the recognition accuracy of acoustic signals degraded by noise. Respecting motion events, this thesis introduces a method based on a root mean square (RMS) framing window in combination with a rotation matrix to achieve robust shake detection and vertical holding detection, respectively. This thesis further describes a series of experiments. Experiment 1 to 3 was designed to characterise the strengths and weaknesses of the system, in which testing also provided useful information for further system development. The final experiment was a real-patient study that involved the recruitment of six patients diagnosed asthma or COPD. The system tracked their pMDI use for three weeks and established the deviations of pMDI usage and technique that existed between the patients' self-reports and independent electronic reports. Characterisation of these differences, and the ability of the system to not only to monitor adherence but also to identify usage failure points suggest that the monitoring system will, in the future, be a highly desirable component of digital healthcare provision.