rushabhdharia / speechtotext Goto Github PK

1. In Data Generator add input_length and label_length to input (X)
2. In class ASR model move ctc batch cost to init method to remove error
3. If everything works perfectly try different approaches (add attention)

1. solve Cudnn GRU issue (can use GRU cells to stop Cudnn script)
2. Add attention model
3. Add model checkpoint/save/load
4. Explain the code in Speech2Text.ipynb

1. Add Predict Function
2. Check input/label length and ctc_decode issue
3. Save plot on cloud instead of show

1. Improve Model
2. Add Predict function
3. Save Predictions to File
4. Add plots
5. Add comments and improve README

Getting the following error - ValueError: could not broadcast input array from shape (20,583) into shape (20)
Is there any way to bypass this without appending zeros?

1. Add Validation and test
2. Explain the code in Speech2Text.ipynb
3. Check why cost is not minimizing

High Priority

1. Make Data Loader efficient (reduce batch size - maybe)

Low Priority

1. Predict function - accepts a wav file and predicts something
2. Saving plots on speech2text.py