Predict speech recognition model with an sample input

Question

I had trained a speech recognition model with TensorFlow. got a checkpoint named "model.ckpt-1300.meta". Now I want to predict the result with a sample.wav file. Can anyone help?

Answer 1

There are different ways to use the saved checkpoint to build a model and perform a prediction; one of them can be as follow:

• You can use an Estimator with model_fn being the original model function, and warm start the model using the saved checkpoint.
• Then you can creating a custom input function to feed your sample.wav file into the Estimator and run the Estimator to provide predictions.