harshalrohit/deepspeech2-prediction

DeepSpeech2 Prediction

This repository demos how to do prediction using DeepSpeech2. The process followed by the one deployed on server uses CNN_RNN.py to perform both training and prediction (may be not).

This repository is just the simpler version of it, without any complexities of training procedure.

Setup

• Install Conda environment.
• use the included environment.yml file to create the conda environment.

Note: It may have some unnecessary packages as well, but those can be ignored for time-being.

Usage

In prediction.ipynb file, change the directory path in PROJECT_ROOT_DIR appropriately and execute the notebook.

Directory structure

data

• Contains the .wav audio files.
• DeepSpeech2 training procedure requires a fixed directory, but that has been relaxed for prediction process

ml-models

• The subdirectories contain the model (only weights) and preprocessing objects as well.

src

• Contains custom files developed for DeepSpeech2

Additional Info

I have used vscode, so few things may vary for the notebook environment.