RESULTS 921 Bytes
Newer Older
1 2 3 4 5 6 7 8 9
exp/mono0a/decode/cer_9:%WER 80.54 [ 45228 / 56154, 1435 ins, 11484 del, 32309 sub ]
exp/tri1/decode/cer_12:%WER 60.34 [ 33881 / 56154, 2720 ins, 6019 del, 25142 sub ]
exp/tri2/decode/cer_12:%WER 59.69 [ 33521 / 56154, 2800 ins, 5618 del, 25103 sub ]
exp/tri3a/decode/cer_13:%WER 57.65 [ 32370 / 56154, 2535 ins, 5673 del, 24162 sub ]
exp/tri4a/decode/cer_12:%WER 53.02 [ 29774 / 56154, 2724 ins, 4791 del, 22259 sub ]
exp/tri5a/decode/cer_13:%WER 49.67 [ 27891 / 56154, 2877 ins, 4538 del, 20476 sub ]
exp/tri5a_mce/decode/cer_11:%WER 44.74 [ 25125 / 56154, 2112 ins, 4108 del, 18905 sub ]
exp/tri5a_mmi_b0.1/decode/cer_11:%WER 44.24 [ 24840 / 56154, 2060 ins, 4118 del, 18662 sub ]
exp/tri5a_mpe/decode/cer_12:%WER 44.96 [ 25247 / 56154, 2233 ins, 4174 del, 18840 sub ]
naxingyu's avatar
naxingyu committed
10 11 12

# ConvNet with 2 convolutional layers and 2 ReLU layers
exp/nnet2_convnet/decode/cer_10:%WER 40.73 [ 22873 / 56154, 2609 ins, 3712 del, 16552 sub ]