Predicted text is always same about bangalasr HOT 7 CLOSED

@skfaysal1997, Could you please share the notebook that you run for inference.
I'm getting correct output by running inference notebook
Thanks

from bangalasr.

Sure!
https://github.com/MdFaysal97/github-share/blob/main/inference.ipynb

from bangalasr.

@menon92 Here's the inference notebook:
https://nbviewer.org/github/MdFaysal97/github-share/blob/main/inference.ipynb

from bangalasr.

@MdFaysal97 @skfaysal1997 Do you use the model that provided in this repository ? or you use your custom model ?

from bangalasr.

@menon92 Here's is the inference output of the model provided in the repo for my custom data.
https://nbviewer.org/github/MdFaysal97/github-share/blob/main/asr_inferance.ipynb
I thought the miss prediction make sense as you have already told it's a biased model. But when i tried training the model using your script i got the following error:

Anyway we can ignore the error by not passing the "display_cb" to the callbacks and complete the training.After training while i inference the model i got prediction like:
https://github.com/menon92/BangalASR/issues/2#issue-1018226133

For your convenient,The scripts are:
train.py
https://gist.github.com/MdFaysal97/34e1c03bbe8a2025cb7d2a185fe399fa
utils.py
https://gist.github.com/MdFaysal97/8c034076fd7d2467f13dbeaf1e1284db
config.py
https://gist.github.com/MdFaysal97/06bf4f1027baa167e1ba0b841d8962a6
Can you please help?

from bangalasr.

I am also facing the same issue. I am using the model provided in this repo

from bangalasr.

@MdFaysal97 , @rayat927 The was a bug in training code.
I did not pass the unique chars to VectorizeChar class, I update the code

vectorizer = VectorizeChar(unique_chars, max_target_len)

Thanks

from bangalasr.