Table 3 Comparison of matching accuracies using the extracted pitch data by the STA pitch extractor with the 2006 MIREX QBSH corpus database
From: Intelligent query by humming system based on score level fusion of multiple classifiers
Method | Criterion | ||||
|---|---|---|---|---|---|
| Â | Â | Top 1 (%) | Top 10 (%) | Top 20 (%) | MRR |
LS (Section 2.4.1) | 52.31 | 77.62 | 89.45 | 0.610 | |
Pitch-based DTW (Section 2.4.2) | 51.76 | 79.75 | 89.42 | 0.605 | |
QB code-based LS (Section 2.4.3) | 59.35 | 78.94 | 89.20 | 0.664 | |
Feature point overlap (Section 2.4.4) | 37.61 | 68.22 | 79.26 | 0.483 | |
Distribution (Section 2.4.5) | 10.55 | 53.49 | 76.15 | 0.239 | |
Decision level fusion (OR) | 46.13 | 75.03 | 98.56 | 0.573 | |
Previous method [13] | 65.40 | 82.37 | 91.12 | 0.702 | |
Score level fusion | MIN | 72.08 | 79.78 | 88.56 | 0.747 |
| Â | MAX | 58.51 | 81.12 | 88.46 | 0.665 |
| Â | SUM | 63.15 | 81.29 | 90.32 | 0.693 |
| Â | Weighted SUM | 62.68 | 81.11 | 90.46 | 0.690 |
| Â | PRODUCT (proposed method) | 72.98 | 81.50 | 90.32 | 0.754 |