VocalWell - Voice Pathology Diagnosis
AI-powered speech analysis system for early detection of vocal disorders using deep learning and acoustic feature analysis.
Project Overview
VocalWell addresses the critical issue of late-stage diagnosis of vocal disorders by engineering a sophisticated speech analysis AI using TensorFlow, Librosa, and advanced acoustic processing techniques.
The system performs comprehensive ablation studies on acoustic features including MFCCs, formants, and spectrograms to identify the most predictive indicators of vocal pathology, achieving superior accuracy compared to existing tools.
Optimized for on-device inference with processing times under 10 seconds to ensure patient privacy and enable real-time clinical applications without requiring cloud connectivity.
Key Features
- Advanced acoustic feature extraction using MFCCs, formants, and spectrograms
- Deep learning models trained on comprehensive vocal pathology datasets
- Real-time on-device inference optimized for privacy and speed
- Comprehensive ablation studies validating feature importance
- 7% accuracy improvement over existing open-source acoustic analysis tools
- Open-source preprocessing scripts and evaluation pipeline for reproducibility
Technologies Used
Project Details
Client
Personal Project
Timeline
3 months (2024)
Role
AI/ML Researcher & Developer
© 2025 Jane Doe. All rights reserved.