Analysis of Voice Parameters During Atrial Fibrillation

Background:Atrial fibrillation (AF) may be present without the patient being aware of any symptoms, leading to a delay in or failure of diagnosis. Voice analysis has been used as a monitoring tool for various medical conditions and human speech has been correlated with heart rate. In this study, we examined the feasibility of using voice analysis to identify AF.

Methods: Patients with AF, who were admitted to the Rabin Medical Center for the purpose of cardioversion, participated in the study. Before and after cardioversion, pre-specified vowels and words along with a simultaneous ECG tracing were recorded. Relevant medical and demographic data were noted. A proprietary “speaker dependent” algorithm was used to diagnose a patient’s heart rhythm from his or her speech by learning specific voice parameters from a training set performed during SR. The changes in these parameters during AF were then analyzed in order to differentiate between AF & SR for a specific speaker.

Results: A total of 41 patients were recorded in both AF & SR. Of the 41 patients recorded, 34 were included in the analysis, and seven were not included due to poor voice signals. Using selectable thresholds, the system demonstrated a sensitivity of 64.5 to 94% and a specificity of 80 to 97% in identifying AF.

Conclusion: This study demonstrated the feasibility of voice analysis in the diagnosis of atrial fibrillation. The voice analysis was able to distinguish atrial fibrillation from sinus rhythm in patients undergoing cardioversion.