14-03-16 Zhiyong Zhang
From cslt Wiki
1, Sinovoice 6000h mpe training has been started; 2, Finished gmm-dnn co-training based on wsj datasets; 3, 6000h's alis and denlats using 1700h 5.nnet model has been prepared for mpe training; 4, Finished decoding test on Sinovoice testcase by add a minus insertion penalty to L's loop and sil state .
Target of next week:
1, Continue the gmm-dnn co-training based on Tencent dataset; 2, Start Sinovoice 6000h mpe training based on 1700h_mpe_5.nnet model's alis and denlats;