Face V/S Fake: Real-Time Deep fake Detection Application for Android Devices

Abstract

Parkinson’s disease (PD) is a progressive neuro- logical disorder that primarily impacts movement, resulting in symptoms such as tremors, stiffness,slowness and balance issues. Since early interventionmay still be effective in delaying disease progression and improving care and quality of life, timely diagnosis is crucial. Diagnosis Despite being aware of the early symptoms, routine diagnosis relies on neurologists’ clinical examinations – something which can belengthy and subjective, also requiring patients to visit hospital multiple times. Strangely enough, slight changes in tone and tremor or raspy delivery that you’ll often overlook when you’re chatting are some of the earliest signs of Parkinson’s disease.For the early diagnosis of Parkinson’s disease, this work explores a voice-enabled machine learning system. The system identifies patterns indicative of Parkinson’s disease through analysis of acoustic features in speech samples. Parameters such as pitch, jitter, shimmer, harmonics-to-noise ratio and other percepts that measure the lack of stability in a person’s voice due to disease are key features. To tackle these inconsistencies, the system leverages a dataset of voice recordings taken both from people with Parkinson’s and those without, in order to normalise values and prepare it for training strong machine learning models.These voice features are then used to train a number of machine learning algorithms, such as Random Forest, Support Vector Machines (SVM), Logistic Regression. The robustness of our models is guaranteed by extensive optimization via cross- validation and hyperparameter tuning. Among them, Random Forest stands out by its accuracy as well as interpretability as we are able to know which part of the voice contributes the most for predicting. The high recognition performance of the system to discriminate PD patients from healthy subjects is confirmed according to evaluation metrics such as accuracy, precision, recallF1-score and ROC-AUC.Streamlit is employed to integrate the trained model into an accessible web interface. The users have three alternative ways to engage with the system, by: manually entering features,uploading a pre-extracted feature CSV file, or directly recording their voice. Besides informing about the relevant voice features that influence the choice,such system also provides an immediate prediction as to whether or not some has Parkinson’s. Doctors could have an additional, noninvasive tool to help them diagnose patients early in an easy and simple way through this method that allows screening and following the patient without having to makenumerous visits to the hospital.

In conclusion, this project offers an approachable, compre- hensible, and efficient Parkinson’s disease detection solution by fusing digital signal processing, artificial intelligence, and web technology.