Key points are not available for this paper at this time.
We present the first open challenge on predicting lyric intelligibility. A new dataset, CLIP1, was introduced, comprising audio samples of popular western music paired with listener intelligibility scores. To model diverse listening profiles, samples were processed with no, mild and moderate simulated hearing loss. A total of 27 systems were submitted by 22 teams. Most systems used foundation models to extract encoder embeddings as high-level acoustic representations, often complemented by signal features and perceptual metrics. Twenty-five systems outperformed the STOI baseline, and 16 outperformed a Whisper-based baseline.
Roa-Dabike et al. (Tue,) studied this question.