Spec2Vec is a Python framework that turns any 1D time-series or 2D time-frequency representation into a compact, interpretable “spectral fingerprint” — a vector of information-theoretic, statistical, and spatial metrics that uniquely characterizes a signal. Developed for geophysics and planetary exploration, it provides a transparent, physics-aware alternative to black-box deep learning for signal characterization and clustering. You can apply it to raw waveforms, STFT spectrograms, or continuous wavelet transforms with just a few lines of Python.
Get Started
Introduction
Learn what Spec2Vec does, who it is for, and how it differs from black-box deep learning approaches.
Installation
Set up your Python environment and install all required dependencies using conda or pip.
Quickstart
Extract your first spectral fingerprint from a synthetic signal in just a few lines of Python.
Concepts Overview
Understand the pipeline architecture — pointwise metrics, pairwise metrics, and Hilbert curve encoding.
Applications
Distributed Acoustic Sensing
Process and characterize fiber-optic DAS arrays using spectral fingerprints.
ESEC Catalog Clustering
Cluster tectonic and non-tectonic earthquakes by their textural and statistical properties.
Synthetic Signal Clustering
Generate and cluster randomized geophysical waveforms using time-frequency fingerprints.
Acoustic Signal Modeling
Model waveforms propagating through heterogeneous media and distinguish them with Spec2Vec.
How It Works
1
Install Spec2Vec
Clone the repository and install dependencies into a fresh conda environment with Python 3.10.
2
Load Your Data
Pass your 1D time-series or 2D spectrogram array directly to the
GISQAPipeline — no special data format required.3
Extract Spectral Features
Call
compute_pointwise_metrics_from_spec() to receive a pandas DataFrame of interpretable spectral features ready for clustering, classification, or visualization.