WSJ
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
WSJ (92-eval)
1.3WER
131
WSJ (test)
95.74UAS
67
WSJ nov93 (dev)
3WER
52
WSJ (test)
97.78Accuracy
51
WSJ (test)
1.2WER
35
WSJ nov92 (test)
1.42WER
34
WSJ (test)
84.3Max F1
29
WSJ Penn Treebank (test)
95.84F1 Score
27
WSJ 10 or fewer words (test)
79.9UAS
25
80-hour WSJ (dev93)
5.7WER
16
WSJ section 23 (all lengths) (test)
65.8Directed Dependency Accuracy (DDA)
16
WSJ section 23 length <= 10 (test)
77.2DDA
16
WSJ0 UNI
3.15PESQ
15
WSJ0-2mix
24ΔSI-SDR (dB)
13
WSJ entire corpus (full)
80.8M1 Score
13
WSJ (test)
98.38PRA
13
WSJ 93 (test)
4.98WER
13
WSJ (test)
0.809AP
12
WSJ (test)
0.01WER
12
WSJ (test)
84.3F1 Score
11
WSJ (dev)
97.37Accuracy
11
WSJ section 23 (test)
96.2UAS
10
WSJ permuted document (test)
98.59Accuracy
8
WSJ word-level gold trees (test)
54.08F1
8
WSJ Penn Treebank (test)
0.727Compatibility (%) - All
7