Description of your solution

Our Project Mental Wellness Companion,

focuses on early detection of depression and stress levels through speech analysis. Mental health is one of the most pressing challenges of our time, and many people avoid seeking help until late. We aim to create a non-intrusive, AI-driven tool that can analyze a person’s voice and detect stress indicators in real-time.

How it works:

1. We use Librosa to extract MFCC (Mel-Frequency Cepstral Coefficients) and related audio features from speech samples. These features capture tone, pitch, and rhythm patterns that are strongly correlated with emotional states.

2. We preprocess all audio files into fixed-length numeric vectors, ensuring consistent input for the model.

3. An XGBoost classifier is trained on these features, achieving ~77% accuracy in distinguishing stressed vs. non-stressed speech samples.

4. We have developed a prototype UI with Gradio, where users can upload or record their voice and receive instant predictions.

What we plan to update:

Improve accuracy further by adding delta & delta² features, more MFCCs, and hyperparameter tuning.

Extend the prototype into a fully functional website hosted on Hugging Face Spaces, making it accessible to anyone with just a browser.

Add additional visualizations (confidence scores, stress patterns, and emotion breakdowns).

Integrate real-time microphone input for more natural user interaction.

Impact:

Early screening of stress and depression in schools, workplaces, and telemedicine platforms.

Non-intrusive monitoring for patients undergoing therapy.

Scalable, privacy-preserving, and deployable on the web.

Currently, our Gradio prototype demonstrates the feasibility, and we will be updating the project into a working Hugging Face-hosted website for the next phase.