Investigation of Deepfake Voice Detection Using Speech Pause Patterns: Algorithm Development and Validation.

BACKGROUND: The digital era has witnessed an escalating dependence on digital platforms for news and information, coupled with the advent of "deepfake" technology. Deepfakes, leveraging deep learning models on extensive data sets of voice recordings and images, pose substantial threats to media authenticity, potentially leading to unethical misuse such as impersonation and the dissemination of false information. OBJECTIVE: To counteract this challenge, this study aims to introduce the concept of innate biological processes to discern between authentic human voices and cloned voices. We propose that the presence or absence of certain perceptual features, such as pauses in speech, can effectively distinguish between cloned and authentic audio. METHODS: A total of 49 adult participants representing diverse ethnic backgrounds and accents were recruited. Each participant contributed voice samples for the training of up to 3 distinct voice cloning text-to-speech models and 3 control paragraphs. Subsequently, the cloning models generated synthetic versions of the control paragraphs, resulting in a data set consisting of up to 9 cloned audio samples and 3 control samples per participant. We analyzed the speech pauses caused by biological actions such as respiration, swallowing, and cognitive processes. Five audio features corresponding to speech pause profiles were calculated. Differences between authentic and cloned audio for these features were assessed, and 5 classical machine learning algorithms were implemented using these features to create a prediction model. The generalization capability of the optimal model was evaluated through testing on unseen data, incorporating a model-naive generator, a model-naive paragraph, and model-naive participants. RESULTS: Cloned audio exhibited significantly increased time between pauses (P<.001), decreased variation in speech segment length (P=.003), increased overall proportion of time speaking (P=.04), and decreased rates of micro- and macropauses in speech (both P=.01). Five machine learning models were implemented using these features, with the AdaBoost model demonstrating the highest performance, achieving a 5-fold cross-validation balanced accuracy of 0.81 (SD 0.05). Other models included support vector machine (balanced accuracy 0.79, SD 0.03), random forest (balanced accuracy 0.78, SD 0.04), logistic regression, and decision tree (balanced accuracies 0.76, SD 0.10 and 0.72, SD 0.06). When evaluating the optimal AdaBoost model, it achieved an overall test accuracy of 0.79 when predicting unseen data. CONCLUSIONS: The incorporation of perceptual, biological features into machine learning models demonstrates promising results in distinguishing between authentic human voices and cloned audio.

An In-Silico model for evaluating the directional shock vectors in terminating and modulating rotors.

Out-of-hospital cardiac arrest (OHCA) accounts for a majority of mortality worldwide. Survivability from an OHCA highly depends on timely and effective defibrillation. Most of the OHCA cases are due to ventricular fibrillation (VF), a lethal form of cardiac arrhythmia. During VF, previous studies have shown the presence of spatiotemporally organized electrical activities called rotors and that terminating these rotor-like activities could modulate or terminate VF in an in-hospital or research setting. However, such an approach is not feasible for OHCA scenarios. In the case of an OHCA, external defibrillation remains the main therapeutic option despite the low survival rates. In this study, we evaluated whether defibrillation effectiveness in an OHCA scenario could be improved if a shock vector directly targets rotor-like, spatiotemporal electrical activities on the myocardium. Specifically, we hypothesized that the position of defibrillator pads with respect to a rotor's core axis and shock current density could influence the likelihood of rotor termination and thereby result in successful defibrillation. We created a bidomain cardiac model based on porcine heart data using Aliev-Panfilov bidomain equations. We simulated localized rotors, which we attempted to terminate using different defibrillation pad orientations relative to the rotor axis (i.e., perpendicular, parallel, and oblique). In addition, we gradually increased current densities for each defibrillation pad orientation from 4 to 12 A/m2. We repeated the above defibrillation procedure for rotors originating from four different locations on the ventricles. The shock parameters and the outcomes were analyzed using a Generalized Linear Mixed Model (GLMM) with Logistic Regression to link rotor termination with the defibrillation pad orientation and current density. Our results suggest the highest average likelihood of terminating rotors during VF is when defibrillator pads are placed perpendicular to the rotor axis (0.99 ± 0.03), with an average current density of 7.2 A/m2, compared to any other orientation (parallel: 0.76 ± 0.26 and oblique: 0.08 ± 0.12). Our simulations suggest that optimal defibrillator pad orientation, combined with sufficient current density magnitude, could improve the likelihood of rotor termination during VF and thereby improving defibrillation success in OHCA patients.

Effect of Shock Vector Orientation in Modulating and Terminating Rotors - a Simulation Study.

The main treatment option for Ventricular Fibrillation (VF), especially in out-of-hospital cardiac arrests (OHCA) is defibrillation. Typically, the survival-to-discharge rates are very poor for OHCA. Existing studies have shown that rotors may be the sources of arrhythmia and ablating them could modulate or terminate VF. However, tracking rotors and ablating them is not a feasible solution in a OHCA scenario. Hence, if the sources (or rotors) can be regionally localized non-invasively and this information can be used to direct the orientation of the shock vectors, it may aid the termination of rotors and defibrillation success. In this work, using computational modeling, we present our initial results on testing the effect of shock vector orientation on modulating (or) terminating rotors. A combination of Sovilj's and Aliev Panfilov's monodomain cardiac models were used in inducing rotors and testing the effect of shock vector magnitude and direction. Based on our simulation results on an average with four experimental trials, a shock vector directed in the perpendicular direction along the axis of the rotor terminated the rotor with 16% lesser magnitude than parallel direction and 38% lesser magnitude than in oblique direction.Clinical Relevance- A rotor localization dependent defibrillation strategy may aid the defibrillation protocol procedures to improve the survival rates. Based on the four experimental trials, the results indicate shock vectors oriented perpendicular to the axis of the rotors were efficient in modulating or terminating rotors with lower magnitude than other directions.