Tag
The paper introduces GALA, a two-stage method for text-to-time-series synthesis that uses generation-aware cross-modal alignment to achieve state-of-the-art results on the TSFragment-600K benchmark, improving both fidelity and caption adherence.