Evaluating the Potential of Vocal Features as Non-Invasive Biomarkers for Heart Failure

This paper explores the potential of vocal features as non-invasive biomarkers for heart failure detection. Changes in voice may result from disruptions in the respiratory and cardiovascular systems, making voice a promising clinical signal. A standardized recording protocol was defined, consisting of five tasks: (1) reading a text at a normal speaking pace, (2) free speech for at least 30 seconds, (3) counting from 1 to 30, (4) counting down from 30 to 1, and (5) a maximum phonation task ("aaa" for as long as possible). From these recordings, various acoustic features are extracted, including phonation duration, number and length of pauses, pitch, loudness, mel-frequency cepstral coefficients (MFCCs), jitter, shimmer, and cepstral peak prominence (CPP). Clinical studies are currently underway to assess the effectiveness of this methodology and its applicability in real-world settings as a tool for early, non-invasive diagnosis and monitoring of heart failure. This approach could empower patient self-management and help reduce healthcare costs. This work was presented at the ETRAN/IcETRAN 2026 conference, Srebrno jezero (Silver Lake), Serbia, and was carried out within the STRATIFYHF project.