ASACA 0.1.3

Automatic Speech Analysis for Cognitive Assessment

ASACA – Automatic Speech Analysis for Cognitive Assessments

ASACA is an end-to-end toolkit that transforms raw speech into multimodal biomarkers — lexical, prosodic and pause-based — and returns an interpretable prediction ( HC / MCI / AD ) and low Word error rate transcriptions (WER <0.02)).

---

✨ Key Features

Capability	Detail
Single-command inference	asaca run audio.wav outputs JSON + PDF report
Fine-tuned wav2vec 2.0 ASR	< 2 % WER on in-domain test set
Explainability	SHAP plots per classification
Rich feature set	word-error rate, syllable rate, pause stats, spectral cues
Offline-ready	Model weights stored under Models/ via Git LFS
PEP 517/621 packaging	pip install asaca or editable mode

---

🚀 Quick start

Install the package from PyPI and run inference on a WAV file:

pip install asaca
asaca-cli gui

Alternatively install in editable mode for development:

git clone https://github.com/RhysonYang-2030/ASACA-Automatic-Speech-Analysis-for-Cognitive-Assessment.git
cd ASACA-Automatic-Speech-Analysis-for-Cognitive-Assessment
pip install -e .[dev]

The CLI outputs recognised text along with a PDF report and JSON file in the specified output directory.

Usage

asaca/
├── src/             # library code
├── tests/           # unit tests
├── docs/            # MkDocs documentation
├── examples/        # example notebooks and data
└── notebooks/       # tutorial notebooks

Run asaca-cli --help to see all commands including feature extraction.

Documentation

Full API reference and user guide live in the docs/ directory and on Read the Docs.

Docker

Build the image and run inference in an isolated environment:

docker build -t asaca .
docker run --rm -v "$PWD:/data" asaca asaca-cli infer /data/audio.wav

The container entrypoint defaults to asaca-cli.

License

Released under the Apache-2.0 license.

Citation

If you use ASACA in your research, please cite the project using the CITATION.cff file.

Contact

Maintainer: Xinbo Yang