Soft Dynamic Time Warping With Variable Step Weights

Johannes Zeitler, Michael Krause, and Meinard Müller

Code for the paper "Soft Dynamic Time Warping With Variable Step Weights", ICASSP 2024

Overview

This repository contains code for using weighted SDTW as a loss function in pytorch. We provide a class for the loss function in <weightedSDTW.py>, and a demo notebook that illustrates the usage <weightedSDTW_demo.ipynb>.