ASR errors correction

Open Challenge for Correcting Errors of Speech Recognition Systems [ver. 1.0.0]

# submitter when ver. description test-A CharMatch test-A Accuracy test-A WER
2 [anonymized] 2020-06-05 20:45 1.0.0 smt on whole words mert=WER 0.1911226221309279 0.683 5.025824394119984e-2
3 [anonymized] 2020-06-05 16:18 1.0.0 Transformer + subword learning batch_size=4096 dropout=0.1 label_smoothing=0.1 layers=6 learning-rate=changing optim=adam steps=80000 warmup_steps=8000 neural-network transformer 0.10249453613686037 0.671 5.9263673685604555e-2
5 [anonymized] 2020-06-05 16:11 1.0.0 This my brilliant solution batch_size=4096 dropout=0.1 label_smoothing=0.1 layers=6 learning-rate=changing optim=adam steps=80000 warmup_steps=8000 neural-network transformer left-to-right 8.50892984342015e-3 0.0 1.0086743477685074
4 [anonymized] 2020-05-27 13:31 1.0.0 dumb way - first try dumz20-challenge 3.980171509208669e-3 8.0e-3 0.38431995762150706
1 [anonymized] 2020-03-17 01:12 1.0.0 1:1, dumbest way possible ^^ stupid dumz20-challenge 0.0 0.752 4.456363395576745e-2