jackaduma / las_mandarin_pytorch Goto Github PK

I got "CUDNN_STATUS_NOT_INITIALIZED"while training the ASR with CTC,Is there a solution? thanks in advance!!

python main.py --config ./config/aishell_asr_example_lstm4atthead1.yaml --test

Using test mode

Traceback (most recent call last):
File "main.py", line 83, in
solver = Solver(config, paras, mode)
File "/data02/sam.yi/LAS_Mandarin_PyTorch/test_asr.py", line 24, in init
super().init(config, paras, mode)
File "/data02/sam.yi/LAS_Mandarin_PyTorch/core/solver.py", line 75, in init
open(config['core']['config'], 'r'), Loader=yaml.FullLoader)
KeyError: 'core'

Hello,

Not be able to open google drive. Can you please share the model in another way?

Hi there is no file named infer.py for inference. Can you add that ?

Firstly, thanks for your share! I've been looking for an implement on ASR using RNNs+Attention/Transformer, just like the translation implements.
Could you please tell us your best error rate using these codes you uploaded? Thanks.

Hello again! I have modified your ASR model struture and used a larger madarin dataset to train it. The CER has already reached down to 10.5 after 1m step without Language Model.

I'd like to see how further it improves by using the LM. But I don't know how to train on your LM codes yet. Could you please tell us how to train it, infer with it, or share your "yaml" file to us? Thanks in advance!