We're excited to announce the first XACLE Challenge (x-to-audio alignment challenge)! This year, we focus on predicting audio-text alignment. While x-to-audio generation (audio from text, video, etc.) is a highly active field, existing objective evaluation methods for output fidelity often show low correlation with human subjective evaluations. This challenge aims to develop an automated model for audio-text alignment prediction that strongly correlates with human subjective evaluations. Our ultimate aim is to faithfully generate audio from human instructions, and evaluating input‒output alignment is crucial for this advancement. Moreover, the development of automated evaluation methods that are strongly correlated with human evaluations is helpful for understanding human audio perception. We warmly welcome researchers from both academia and industry to participate.
