Objective: Aimed to develop a machine learning model for analyzing audio samples to assess the likelihood of various speech defects in children and to integrate this model into a service for use in a mobile app.
Performed comprehensive audio sample preparation using Librosa, SciPy, and NumPy, creating around 50 distinct features. Subsequently, samples were segmented by defect type and clustered to identify similar areas.
Developed a linear classification framework, with a separate model for each defect and an overarching Bayesian model for enhanced analysis.
Fine-tuned each model to achieve precise defect detection, optimizing thresholds and attaining accuracy levels ranging from 70% to 95% across various defect categories.
Integration & Outcome: Successfully integrated the model into the client's application through a robust API implementation, leveraging NumPy for increased processing speed.
Like this project
0
Posted Jan 14, 2024
Built ML model for speech defect analysis in children's audio samples, achieving 70-95% accuracy and seamless integration into a mobile app.