shiyuzh2007 / asr Goto Github PK

Thanks for your wonderful work.
And I want to know the format of the two files,are they the same like kaldi?
I use the kaldi feat.scp and when process the ark file,it's broken.
Thanks.

Hi, thanks for your effort, Is there any pretrained model, trained by you that we can use For Chinese ASR?

I find in your yaml files, you used "tensor2tensor/hkust_ci_phone/src_data/words_s2s.txt.bpe_5000" or something like "hkust_ci_phone/src_data/train_dim80/text.sp.bpe_5000", I want to know how to get this processed data

你好，请问可以看到yaml中所用到的数据格式例子吗？

Hi,
我尝试复现HKUST实验，配置文件用的是：config_template_char_unit512_block6_left3_big_dim80_sp.yaml

我的实验环境受限，只能用3块GPU卡（TITAN X 12G)。
总出现OOM问题，我试着把bucket_step和tokens_per_batch调为默认值的1/3甚至更小，但还是有OOM问题。
请问有任何建议吗？

Would you like the share the download link of the HKUST dataset? Thanks a lot.

在训练一段时间后损失函数无法下降，在预测时出来的都是同一个字，想请问您与有没有出现过这种情况？