BUT QUESST 2015 System Description
All our systems are based on Dynamic Time Warping (DTW). These systems use bottle-neck features (BN) as input. The bottle-neck feature extractors were trained on GlobalPhone Czech, Portuguese, Russian and Spanish languages, so our approach is in low-resource category. We also aimed on T1/T2/T3 types of query search for late submission systems. System calibration and fusion were based on binary logistic regression.
PDF
Datasets
Add Datasets
introduced or used in this paper
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Benchmark |
|---|---|---|---|---|---|---|
| Keyword Spotting | QUESST | BUT [l-fea stack DTW+slope+2w3w fusion] (dev) | Cnxe | 0.8731 | # 24 | |
| MinCnxe | 0.8321 | # 28 | ||||
| Keyword Spotting | QUESST | BUT [l-fea stack DTW 3w+slope] (dev) | Cnxe | 0.9188 | # 31 | |
| MinCnxe | 0.8801 | # 40 | ||||
| Keyword Spotting | QUESST | BUT [l-fea stack DTW 2w+slope] (dev) | Cnxe | 0.8884 | # 27 | |
| MinCnxe | 0.8569 | # 35 | ||||
| Keyword Spotting | QUESST | BUT [l-fea stack DTW+slope] (dev) | Cnxe | 0.8772 | # 26 | |
| MinCnxe | 0.8389 | # 30 | ||||
| Keyword Spotting | QUESST | BUT [p-fea stack DTW ] (dev) | Cnxe | 0.8580 | # 22 | |
| MinCnxe | 0.8426 | # 31 | ||||
| Keyword Spotting | QUESST | BUT [l-fea stack DTW+slope+2w3w fusion] (eval) | Cnxe | 0.8447 | # 19 | |
| MinCnxe | 0.8124 | # 25 | ||||
| Keyword Spotting | QUESST | BUT [l-fea stack DTW+slope] (eval) | Cnxe | 0.8490 | # 21 | |
| MinCnxe | 0.8184 | # 26 | ||||
| Keyword Spotting | QUESST | BUT [p-fea stack DTW ] (eval) | Cnxe | 0.8452 | # 20 | |
| MinCnxe | 0.8263 | # 27 |
Methods
No methods listed for this paper. Add
relevant methods here
