Open gaojingqian opened 4 years ago
where do you get the acoustic scores ? We have provide completed examples in https://github.com/athena-team/athena#5-decoding-with-wfst
I downloaded the athena-decoder fold only.It cannot run correctly follow README.md, and an error occurred that is lack of some txt files. I added these files according to the historical commit version:06a6d95e0e0d8d18aaa7f15f94464eef8fc02bf7. Then, when run this command(following README.md in athena-decoder ), this error happens.
I downloaded the athena-decoder fold only.It cannot run correctly follow README.md, and an error occurred that is lack of some txt files. I added these files according to the historical commit version:06a6d95e0e0d8d18aaa7f15f94464eef8fc02bf7. Then, when run this command(following README.md in athena-decoder ), this error happens.
You need to clone athena and train the acoustic model there first.
To run the whole acoustic training & WFST decoding process, please follow this recipe change 'use_wfst' to true.
Traceback (most recent call last): File "pydecoders/decode_wfst_main.py", line 89, in
decoder.decode(enc_outputs, initial_packed_states, e2e_model.inference_one_step)
File "/algdata03/jingqian.gao/code/fromGit/athena-decoder/pydecoders/decoders/wfst_decoder.py", line 80, in decode
weight_cutoff = self.process_emitting(encoder_outputs, inference_one_step_fn)
File "/algdata03/jingqian.gao/code/fromGit/athena-decoder/pydecoders/decoders/wfst_decoder.py", line 168, in process_emitting
if self.eos == np.argmax(all_log_scores[seq_id]):
IndexError: list index out of range
What can I do?