search

mesolitica / malaya-speech

Speech Toolkit for Malaysian language, https://malaya-speech.readthedocs.io/
https://malaya-speech.readthedocs.io/
MIT License
236 stars 42 forks source link

Retraining of Singlish-Large-Conformer #20

Closed mr-coconut closed 2 years ago

mr-coconut commented 2 years ago

Hi~~~ I am trying to understand the code provided in the repo about how to retrain the conformer model. I am wondering if it is possible to retrain the model from a TensorFlow checkpoint and do you have any tutorial code for it? I am new to TensorFlow and found it hard to grasp the whole idea of how to retrain the model. Thanks so much!

huseinzol05 commented 2 years ago

Pretrained checkpoints at https://github.com/huseinzol05/malaya-speech/tree/master/pretrained-model/stt/conformer, for large singlish conformer is no 24.

Training script at https://github.com/huseinzol05/malaya-speech/blob/master/pretrained-model/stt/conformer/large-singlish.py, to load specific checkpoint, you can follow https://github.com/huseinzol05/malaya-speech/blob/master/pretrained-model/stt/conformer/base.py#L340, from line no 340-348, all this you can read more about Tensorflow Estimator, https://www.tensorflow.org/guide/estimator

My code heavily use Tensorflow 1, you might want upgrade to Tensorflow 2.