Transliteration Model

We have developed a transliteration model @ Xelpmoc which transliterates বাংলা to English, with an accuracy of approximately 96.5 %.

Important Files

• transliteration_model.ipynb - This the Jupyter notebook file which was used to train the transliteration model. The neural network was trained on a Google Cloud Platform instance having 2vCPUs and a single NVIDIA Tesla T4 GPU, taking around 5 minutes for a single epoch.

• transliteration_model_usage.ipynb - This the Jupyter notebook file which was used to develop the API use_model.py. It includes all the necessary preprocessing required to feed the বাংলা word to the API, so as to get the necessary output.

• transliteration_model.h5 - This is the trained Keras Sequential Model which does all the prediction for us.

• use_model.py - This is the API which we have developed, that takes the input through a command line argument, and gives the output in the file output.txt.

• data/ - This folder contains the dataset on which the neural network was trained.

Additional Feature(s)

• machine_translation.ipynb - This Jupyter Notebook contains 5 models which developed a translation model which could convert English to French. The hybrid model is the one on which we based our final transliteration model. [Credit : Tommy Tracey]

Note

• Version of TensorFlow used : 1.11.0
• Version of Keras used : 2.2.4

Usage :

Correct Usage of API : python3 use_model.py -word <bengali_word>
For example : python3 use_model.py -word সরকার