RELATE: Subjective evaluation dataset for automatic evaluation of relevance between text and audio
Sound
2025-07-01 v1 Audio and Speech Processing
Abstract
In text-to-audio (TTA) research, the relevance between input text and output audio is an important evaluation aspect. Traditionally, it has been evaluated from both subjective and objective perspectives. However, subjective evaluation is costly in terms of money and time, and objective evaluation is unclear regarding the correlation to subjective evaluation scores. In this study, we construct RELATE, an open-sourced dataset that subjectively evaluates the relevance. Also, we benchmark a model for automatically predicting the subjective evaluation score from synthesized audio. Our model outperforms a conventional CLAPScore model, and that trend extends to many sound categories.
Cite
@article{arxiv.2506.23582,
title = {RELATE: Subjective evaluation dataset for automatic evaluation of relevance between text and audio},
author = {Yusuke Kanamori and Yuki Okamoto and Taisei Takano and Shinnosuke Takamichi and Yuki Saito and Hiroshi Saruwatari},
journal= {arXiv preprint arXiv:2506.23582},
year = {2025}
}
Comments
Accepted to INTERSPEECH2025