Audio-Text Alignment Scoring on XACLE 2024 (val)
0.678SRCCEnsemble
Evaluation Results
| Method | Links | |
|---|---|---|
| EnsembleConfiguration=Rank averaging of Stage 3 and No AAC variant2026.01 | 0.678 | |
| Stage 1, 2 & 3: XACLE Fine-tuningConfiguration=Full 3-stage training pipeline2026.01 | 0.674 | |
| No AAC Pretraining (Stage 2 & 3 only)Configuration=Pseudo-Label Pretraining and XACLE Fine-tuning2026.01 | 0.669 | |
| HumanCLAP-M2DConfiguration=Teacher model2026.01 | 0.602 | |
| Stage 1 & 2: Pseudo-Label PretrainingConfiguration=AAC and CLAP Pseudo-Label Pretraining2026.01 | 0.598 | |
| No Pretraining (Stage 3 only)Configuration=XACLE Fine-tuning only2026.01 | 0.574 | |
| Official BaselineConfiguration=Default2026.01 | 0.384 | |
| Stage 1: AAC PretrainingConfiguration=AAC Pretraining only, Evaluation Protocol=Cosine similarity of text embeddings2026.01 | 0.352 |