To run the model, first make a data file with each line of the following format:
The end result is the same. :: DH AH _ EH N D _ R IH Z AH L T _ IH Z _ DH AH _ S EY M .
Each line is split by ::, every character in the first part is parsed using
isymbols, and the second part is further split on white spaces
and parsed using osymbols. A sample data file is
included with this repository.
For a larger data file go here.
To train the model, first verify the parameters provided at the beginning of
train.py, and run the following:
$ mkdir ckpts logdir
$ ./train.py
This will run the training procedure in infinite loop, while printing the training progress. I haven't been able to get the accuracy significantly above 30% with the current setup.
While training is in progress, you may run tensorboard --logdir=logdir/ and
see the accuracy and loss values over time.
Once the training is done to your satisfaction, run ./infer and a prompt will
await your input. Enter a line of text and be amazed at its inaccuracy!!