This project aims to detect cognitive decline patterns, such as those associated with dementia, using audio and natural language processing (NLP) features extracted from speech samples. Leveraging the DementiaBank Pitt Corpus, the pipeline processes audio files, extracts features like pause count, speech rate, and lexical diversity, and applies unsupervised learning (K-means and Isolation Forest) to identify potential patterns.

Preprocesses audio files and generates transcripts using faster-whisper.

Extracts audio features (e.g., pause count, speech rate) and NLP features (e.g., hesitation count, lexical diversity) with librosa and spacy.

Applies K-means clustering and Isolation Forest for anomaly detection.

Generates visualizations (e.g., pause count boxplot, speech rate scatter) using matplotlib and seaborn.

Saves results in CSV format and model pickles for further analysis.

Python 3.10

Required packages (install via requirements.txt):

Additional setup:

Clone the repository:

Create a virtual environment:

Install dependencies:

Place audio files (e.g., .wav from DementiaBank Pitt Corpus) in the data/raw_samples/ directory.

Run the pipeline:

Check results in the results/ directory:

src/: Source code (e.g., pipeline.py, feature_extraction.py).

data/: Input data (e.g., raw_samples/, processed/).

results/: Output files (e.g., results.csv, visualizations/).

notebooks/: Jupyter notebooks for analysis (e.g., analysis.ipynb).

requirements.txt: Dependency list.

README.md: This file.

venv/: Virtual environment (optional, add to .gitignore).

The pipeline generates four visualizations:

Pause Count by Cluster: Boxplot of pause frequency.

Speech Rate vs. Pause Count: Scatter plot with cluster hue and risk score size.

Hesitation Count by Cluster: Boxplot of speech disfluencies.

Lexical Diversity vs. Incomplete Sentences: Scatter plot of linguistic features.

Feel free to fork this repository, submit issues, or create pull requests. Suggestions for improving feature extraction or visualization are welcome!

MIT License - See LICENSE for details.

DementiaBank Pitt Corpus for audio data.

For questions or collaboration, reach out to varshith.gaddam@example.com or open an issue on this repository.