This folder provides a working, well-documented example for training a seq2seq (+ CTC) speech recognizer model from scratch, based on a few hours of data.

There are three files here:

• train.py: the main code file, outlines the entire training process.
• train.yaml: the hyperparameters file, sets all parameters of execution.
• mini_librispeech_prepare.py: If necessary, downloads and prepares data manifests.

To train the speech recognition model, just execute the following on the command-line:

python train.py train.yaml

We assume you already trained the tokenizer (see ../Tokenizer) and the language model (../LM). Training is done with the mini-librispeech dataset using a CRDNN model for encoding and a GRU for decoding. We pre-train with a larger model to ensure convergence (mini-librispeech is too small for training an e2e model from scratch). In a real case, you can skip pre-training and train from scratch on a larger dataset.