Early Detection of Alzheimer’s Disease Using Voice Biomarkers and Deep Learning

Abstract

Alzheimer’s disease is a progressive neurodegenerative disorder and the most common cause of dementia worldwide. Early detection plays a vital role in improving treatment effectiveness, patient care planning, and slowing cognitive decline. This research presents an application-based deep learning system designed to detect early-stage Alzheimer’s disease through analysis of voice biomarkers. The solution leverages patient voice recordings to capture linguistic and acoustic markers such as pause duration, speech rate, articulation clarity, pitch variation, amplitude dynamics, and lexical diversity. These features are extracted via automated signal processing pipelines, with categorical inputs (e.g. demographic data) encoded and missing acoustic values imputed using statistically robust methods to maintain dataset consistency. The deep learning architecture combines Convolutional Neural Networks (CNNs) for high resolution spectrogram feature extraction with Long Short-Term Memory (LSTM) networks for temporal sequence modelling, enabling detection of subtle, time-dependent speech pattern changes associated with cognitive decline. The model is rigorously trained, while performance is assessed through accuracy, Area Under the Curve (AUC), F1 score, precision, and recall to ensure reliability and clinical relevance. Unlike web-based deployments, this system is delivered as a standalone, cross-platform desktop application, capable of running locally without internet connectivity. The application includes an intuitive interface for healthcare providers, caregivers, and researchers to record or upload voice samples and receive real-time Alzheimer’s risk assessments directly on their devices.

Introduction

The text describes a deep learning–based system for early detection of Alzheimer’s Disease (AD) using speech biomarkers. Alzheimer’s is a progressive neurodegenerative disorder with no cure, making early detection critical. The system leverages datasets like ADReSS, extracting acoustic features such as Mel-Frequency Cepstral Coefficients (MFCCs) from patient speech. A hybrid CNN-LSTM architecture analyzes spectrograms for spatial patterns and temporal speech sequences, capturing subtle cognitive impairments. The trained model is deployed in a web application where users can upload audio files and receive real-time predictions, including classification, confidence scores, and visualizations.

The approach addresses limitations of conventional diagnostics (MRI, PET) by providing a non-invasive, scalable, and cost-effective tool. Key system components include audio preprocessing (noise reduction, normalization, segmentation), feature extraction, deep learning analysis, and a user-friendly interface. Performance is evaluated using metrics like accuracy, precision, recall, F1-score, and ROC curves. Future enhancements aim at multi-class severity detection, longitudinal tracking, mobile integration, and explainable AI for broader clinical utility.

Conclusion

In Conclusion, A successful implementation of the Early Detection of Alzheimer’s Disease using Voice Biomarkers and Deep Learning demonstrates the potential of AI-driven solutions in healthcare. By leveraging non-invasive speech recordings, the system provides a cost-effective and accessible alternative to traditional diagnostic methods. The project integrates a complete pipeline, beginning with audio preprocessing and feature extraction using Librosa, followed by deep learning model training in Google Colab. The model’s performance is rigorously validated through standard metrics, including Accuracy, Precision, Recall, F1-score, ROC curves, and calibration plots, ensuring reliable predictions and confidence in its results. The structured weekly implementation plan allows systematic development, from dataset acquisition and preprocessing to model training, evaluation, and deployment, ensuring modularity and maintainability. Unlike conventional web projects, this system focuses on machine learning workflows rather than multi-tier architectures or database management, highlighting the advantages of AI-focused applications. Real-time inference, lightweight deployment, and clear visualization of results enhance user experience and accessibility.

References

[1] S. dahiya, S. Vijayalakshmi, and munish Sabharwal, “Alzheimer’s disease detection using machine learning: A review.,” International Conference on Advances in Computing, Communication Control and Networking (ICACCCN), Nov 2022. [2] S. al shoukry and N. M. Makbol, “Alzheimer diseases detection by using deep learning algorithms: A mini-review,” IEEE, May 2020. [3] M. Zaabi, N. Smaoui, H. Derbel, and W. Hariri, “Alzheimer’s disease detection using convolutional neural networks and transfer learning based methods,” IEEE, 2022. [4] Y. Pusparani, P. Ardhiant, I. Farady, J. Sahaya, and R. Alex, “Diagnosis of alzheimer’s disease using convolutional neural network with select slices by landmark on hippocampus in mri images,” IEEE, May 2023. [5] A. Almohimeed, Redhwanm.A.Saad, S. Mostafa, N. Mahmoudel-rashidy, Sarahfarrag, A. Gaballah, and H. Saleh, “Explainable artificial intelligence of multilevel stacking ensemble for detection of alzheimer’s disease based on particle swarm optimization and the sub-scores of cognitive biomarkers,” IEEE, Nov 2023. [6] H. Bohra, D. Diwan, and N. Garg, “Improved alzheimer detection using image enhancement techniques and transfer learning,” IEEE, May 2022. [7] B. A. Chakravarthi and gandlashivakanth, “Integrating multimodal ai techniques and mri preprocessing for enhanced diagnosis of alzheimer’s disease: Clinical applications and research horizons,” IEEE, April 2025. [8] C. Botelho, T. Schultz, and A. Abad, “Speech as a biomarker for disease detection,” IEEE, Nov 2024. [9] Y. F. Khan and B. K. M. Imamrahmani, “Stacked deep dense neural network model to predict alzheimers dementia using audio transcript data,” IEEE, March 2022. [10] Y. F. Khan and B. K. M. Imamrahmani, “Hsi-lfs-bert: Novel hybrid swarm intelligence based linguistics feature selection and computational intelligent model for alzheimers prediction using audio transcript,” IEEE, Nov 2022.

Copyright

Copyright © 2026 Piyush S. Nandre, Harsh J. Patil, Rohan S. Kamble, Shrimat N. Nagarkar, Prof. Mohd. Shakeel Mohd. Iqbal. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.