Edge-Deployed Deep Learning for Real-Time Positive and Negative Affirmation Detection

Tiny Machine Learning (TinyML) enables the deployment of efficient machine learning models on resourceconstrained devices, supporting real-time applications directly at the edge. In this work, we present an audio classification model developed using Edge Impulse to distinguish between positive and negative affirmations. The model was trained and validated on affirmation datasets, achieving an accuracy of 91.4 % with a validation loss of 0.21. The confusion matrix shows$\mathbf{8 8. 5 \%}$accuracy for negative affirmations and$\mathbf{9 4 \%}$accuracy for positive affirmations, with corresponding F1scores of 0.91 and 0.92, indicating robust performance across both classes. This study demonstrates the feasibility of deploying affirmation classification models on low-power devices, paving the way for practical applications in mental health monitoring and support. Potential use cases include integration into therapeutic tools for psychologists, enabling real-time detection of self-talk patterns that may support interventions such as cognitive-behavioural therapy and mood tracking. The results suggest that TinyML can extend psychological support beyond clinical settings into everyday life.