neural_sp

neural_sp copied to clipboard

loss ctc fluctuates

1

hi, I am creating an ASR model on Korean dataset. My E2E ASR model without language model works really well, wer around 1 on dev and evaluation set. But When...

qkrguswn2401

Hi, I'm trying to reproduce the results on librispeech but run into troubles with Kaldi. Can I do data preprocessing with codes from [ data_preproceessing ](https://github.com/hirofumi0810/asr_preprocessing) repo instead of using...

artificertxj1

Need help with instructions to reproduce experiments

10

Hi Hiro! First, thank you for the repo. I've been following for a while and I saw you implement a big number of dl architectures. So far I was only...

jiwidi

What are the recommended versions of Python, Pytorch, and CUDA in order to run examples/librispeech?

SuryodayBasak

Can someone list architectures that support stream decoding?

Durgesh92

How to decode empty target files which only contain background noise

1

I have some background noise wav files in my testset, in order to test the robustness of models. But I find it would remove the empty examples in bin/asr/eval.py when...

CSLujunyu

quetion about the loss and grad of "mbr"

3

https://github.com/hirofumi0810/neural_sp/blob/2b10b9cc4bdecb5180ecc45575c0ef410fb09aa3/neural_sp/models/seq2seq/decoders/las.py#L535-L548 I don't know much about mbr, according to these lines, it looks like a [mWER](https://ieeexplore.ieee.org/abstract/document/8461809) loss and gradient to me

Cescfangs

Hi, In speech tasks, i think the performance of the language model should be correlated with the ASR performance, but There are few ASR performance comparisons between different language models...

Garry-sh

what's version fo cuda and pytorch

2

i run the aishell example(transducer) and use torch1.4+cuda10.1 or torch1.5+cuda10.1, there are some error as follows： torch1.4+cuda10.1: /opt/conda/conda-bld/pytorch_1579022060824/work/aten/src/ATen/native/cudnn/RNN.cpp:1266: UserWarning: RNN module weights are not part of single contiguous chunk of...

jinggaizi

bash run error

2

Traceback (most recent call last): File "/home/qidongchen/xxk/neural_sp/examples/aishell/s5/../../../neural_sp/bin/lm/train.py", line 341, in save_path = pr.runcall(main) File "/home/qidongchen/anaconda3/envs/torch1.15/lib/python3.7/cProfile.py", line 121, in runcall return func(*args, **kw) File "/home/qidongchen/xxk/neural_sp/examples/aishell/s5/../../../neural_sp/bin/lm/train.py", line 176, in main benchmark=not is_transformer...

xiexukang