Respiration rate (RR) and other respiratory features, such as inhale-to-exhale ratio (IER) and duration of breathing phases can be used as marker of respiratory and lung conditions and to modulate autonomic function in biofeedback applications. In this study, audio respiration signals were recorded by 112 participants using smartphones. RR was estimated using a frequency-domain method. Acoustic features were extracted from the audio signals and random forest was used to classify inhales, exhales and respiratory pauses, with ROC AUCs of 0.84 and 0.95 for inhales and exhales respectively. RR was estimated with a mean absolute error (MAE) of 0.63 bpm. IER was estimated with a MAE of 0.37, with 76% of the dataset reporting a MAE of less than 0.20. The results demonstrate a computationally efficient approach to estimate respiratory features from audio signals recorded using smartphones that can be easily implemented in real-time for large-scale home monitoring or biofeedback applications.