Speech Signal Parameters as Biomarkers for Remote Monitoring of Patients with Chronic Heart Failure

Aim    To evaluate changes in speech signal parameters during treatment in patients with chronic heart failure (CHF) and to optimize a set of speech parameters that can be used for remote monitoring of patients' condition after treatment.Material and methods    Speech signals of 55 patients with CHF during exacerbation and 38 patients of the same group during remission were analyzed using a proprietary technique. The results were compared with speech signal data of 57 apparently healthy individuals. The following acoustic and prosodic parameters were calculated for the three groups using the Praat v 6.4.35 software: mean, minimum, and maximum values of the fundamental tone frequency, its standard deviation, variation range and mean absolute slope, jitters (local, abs, rap, ppq5, ddp), shimmers (local, apq3, apq5, apq11, dda), harmonic-to-noise ratio, and the ratio of the number of voiced frames to the total number of frames.Results    The study compared three groups: patients before treatment (group 1.1), after treatment (group 1.2), and a control group of apparently healthy individuals (group 2). Analysis of speech signal parameters showed that patients before treatment had significantly different from the control values of several parameters, which reflected the frequency and amplitude instability of the voice. After the course of therapy, the Jitter (local) value was significantly decreased (p=0.012), while in group 1.2, the jitter values did not differ from the values in the control group, indicating the normalization of the frequency stability of the voice signal. The Bowley skew index also was significantly increased (p=0.041) and approached the values of the control group (p=0.068). The Shimmer (dda) and Shimmer (apq3) indexes did not show positive dynamics and maintained significant differences from the control values.Conclusion    The study showed that during treatment of patients with CHF, as their condition improved, jitter significantly decreased while the nonparametric pitch asymmetry coefficient increased and approached the control values. Other speech parameters either did not change significantly or did not approach the values in the control group. This finding can be used for remote monitoring of CHF patients after hospital discharge.