Explore smaller payloads for storage and transfer, while checking the quality of the reconstructed signal.
Explore use cases →
compressionKITv0.2.0Keep the signal.
Carry less data.
A Python-based AI Development Kit for compressing ECG and PPG signals. Compare codecs, measure reconstructed signal quality, and prepare models for edge applications.
From samples to a codec
- 01Prepare signalsECG and PPG preprocessing
- 02Build codecsNeural, DSP and hybrid methods
- 03Evaluate resultsWaveform and physiological scorecards
- 04Prepare for the edgeModels, manifests and reference vectors
What is compressionKIT?
Section titled “What is compressionKIT?”compressionKIT is a Python development kit for compressing ECG and PPG waveforms. It brings codec models, signal preprocessing, evaluation and export into one workflow, so you can compare approaches before preparing an application for the edge.
A codec encodes a frame of samples into a smaller representation and decodes it back into a waveform. Compression can change the reconstructed signal. The evidence pages help you inspect that tradeoff alongside payload size, rather than relying on compression ratio alone.
What you can do
Section titled “What you can do”Inspect waveform and physiological measures across PPG and ECG configurations.
Review the evidence →
Export models, codebooks, manifests and reference vectors for integration work.
Explore deployment →
Use a provided experiment or compose your own workflow from the available blocks.
Build your own workflow →
Installation
Section titled “Installation”Use Python 3.12 and a source checkout to get started:
git clone https://github.com/AmbiqAI/compressionkit.gitcd compressionkituv sync --python 3.12 --extra hfThe hf extra supports loading remote Hugging Face bundles. The Getting Started guide runs a first encode/decode example; no training dataset is needed for that path.
Explore the documentation
Section titled “Explore the documentation”Install compressionKIT and run your first encode/decode example.
Start here →
Load a remote bundle or an existing local deploy directory.
Load a codec →
Review signal-specific results, configurations and reconstruction plots.
Browse models →
Find a registered configuration, prepare its dataset and run the workflow.
Explore experiments →