This is the unofficial implementation of Attention-Based Models for Speech Recognition. (18660 Optimization project)
We evaluate our implementation on the public dataset TIMIT. Kaldi is adopted to process audio data to filter bank feature.
You can update your training configurations and hyperparameters in the config.py
.
Update the data path in config.py
and run the following code to train the ARSG.
$ python main.py