An Initial Study of Physiological Markers for Deepfake Speech Detection

Authors: Vojtěch Staněk, Michal Frič, Kamil Malinka, Anton Firc

Published: 2026-10-13 11:41:04+00:00

Comment: Accepted at BIOSIG 2026

AI Summary

This study proposes detecting deepfake speech by identifying inconsistencies in physiological markers related to respiration, phonation, and temporal coordination, which synthesizers may not faithfully reproduce. Using 36 handcrafted markers and logistic regression or SVM classifiers, the approach achieves an EER of 21.40% in-domain and shows interpretable signals for distinguishing bona fide from deepfake speech.

Abstract

Conventional deepfake speech detectors primarily target synthesis artifacts and often degrade under unseen attacks or recording conditions. Human speech, however, contains physiological markers associated with respiration, phonation, and temporal coordination arising from the speech-production system that synthesizers may not reproduce faithfully. This motivates the hypothesis that deepfake speech can be detected by identifying inconsistencies in these patterns, which we test using 36 handcrafted physiological markers organized into respiratory, vocal, and temporal categories. We evaluate the marker vectors with logistic regression and SVM classifiers on pooled English recordings. The best in-domain system achieves a 21.40% equal error rate (EER), while logistic regression achieves 32.77% on the out-of-domain dataset pair. Since the markers describe specific speech-production properties, the logistic-regression coefficients indicate whether higher marker values push the classifier toward bona fide or deepfake speech. These results provide initial evidence that physiological markers offer a useful, interpretable signal for deepfake speech detection.

Key findings

The physiological marker-based detector achieved an EER of 21.40% in-domain using an RBF SVM, and logistic regression demonstrated more stable performance across different evaluation settings. The coefficients of the logistic regression model provide intrinsic marker-level explanations, indicating which physiological characteristics contribute to classifying speech as bona fide or deepfake, offering an interpretable signal for detection.

Approach

The proposed method extracts 36 handcrafted physiological markers from speech recordings, categorized into respiratory, vocal, and temporal properties. These markers are then fed into linear classifiers (logistic regression or SVM) to predict whether the speech is bona fide or deepfake. The interpretability of the linear model's coefficients allows for identifying which specific physiological markers contribute to the classification.

Datasets

EDS-14 Pooled Dataset (combining ASVspoof 5, Codecfake, In-the-Wild, M-AILABS, MLAAD, SCDF, SpeechFake)

Model(s)

Logistic Regression (LR), Support Vector Machine (SVM) with Radial Basis Function (RBF) kernel

Author countries

Czech Republic

← Previous