Comments (4)
@Nakachi-S Yes. If you want to use cold/deep fusions, you have to stop at stage-3 once.
This is because the LM directory name is defined in python scripts.
from neural_sp.
@hirofumi0810 Thank you for reply!
I resumed ASR training as you suggest, specifying the language model from stage 4.
(The added options, which are Lm_fusion and external_lm, are described above comment .)
Learning the first epoch works well.
But then, the following error occurs during the evaluation.
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▉| 370180/370189 [3:41:03<00:01, 5.61it/s]Traceback (most recent call last):
File "/home/jupyter-nakachi-s/neural_sp/examples/csj/s5/../../../neural_sp/bin/asr/train.py", line 529, in <module>
save_path = pr.runcall(main)
File "/opt/tljh/user/envs/ncr_env/lib/python3.7/cProfile.py", line 121, in runcall
return func(*args, **kw)
File "/home/jupyter-nakachi-s/neural_sp/examples/csj/s5/../../../neural_sp/bin/asr/train.py", line 424, in main
optimizer.n_epochs + 1, logger)
File "/home/jupyter-nakachi-s/neural_sp/examples/csj/s5/../../../neural_sp/bin/asr/train.py", line 481, in evaluate
metric, cer = eval_wordpiece(models, dataset, recog_params, epoch=epoch)
File "/home/jupyter-nakachi-s/neural_sp/neural_sp/evaluators/wordpiece.py", line 82, in eval_wordpiece
ensemble_models=models[1:] if len(models) > 1 else [])
File "/home/jupyter-nakachi-s/neural_sp/neural_sp/models/seq2seq/speech2text.py", line 707, in decode
exclude_eos, refs_id, utt_ids, speakers)
File "/home/jupyter-nakachi-s/neural_sp/neural_sp/models/seq2seq/decoders/las.py", line 911, in greedy
lmout, lmstate = self.lm.decode(self.lm(y), lmstate)
File "/home/jupyter-nakachi-s/.local/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in __call__
result = self.forward(*input, **kwargs)
File "/home/jupyter-nakachi-s/neural_sp/neural_sp/models/lm/lm_base.py", line 62, in forward
loss, state, observation = self._forward(ys, state)
File "/home/jupyter-nakachi-s/neural_sp/neural_sp/models/lm/lm_base.py", line 66, in _forward
ys = [np2tensor(y, self.device_id) for y in ys] # <eos> is included
File "/home/jupyter-nakachi-s/neural_sp/neural_sp/models/lm/lm_base.py", line 66, in <listcomp>
ys = [np2tensor(y, self.device_id) for y in ys] # <eos> is included
File "/home/jupyter-nakachi-s/neural_sp/neural_sp/models/torch_utils.py", line 49, in np2tensor
tensor = torch.from_numpy(array)
TypeError: expected np.ndarray (got Tensor)
Have ever seen like this error?
from neural_sp.
@Nakachi-S Thank you for your report. I fixed this in #87.
Please try the latest code.
from neural_sp.
It works!!!
Thank you your help!
from neural_sp.
Related Issues (20)
- NaN or Inf found in input tensor HOT 3
- Unable to reproduce MoCha with librispeech HOT 4
- error while runing score.sh in aishell HOT 4
- Streaming Transformer Transducer HOT 9
- RNN-T,growing memory occupancy of GPU.
- Issues with score.sh on streaming transformer(mma) models
- examples/ami/s5b recipe failing HOT 2
- question about rnnt result HOT 2
- what's version fo cuda and pytorch HOT 2
- Can provide ASR performance results for different language models?
- Aishell example die in stage 1
- a question about transducer training HOT 1
- conformer transducer HOT 1
- quetion about the loss and grad of "mbr" HOT 3
- How to decode empty target files which only contain background noise HOT 1
- Streamable architecture
- Recommended versions of Python, Pytorch, CUDA?
- data preprocessing
- librispeech dict issue vocab HOT 3
- loss ctc fluctuates HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from neural_sp.