Speech is an open-source package to build end-to-end models for automatic speech recognition. Sequence-to-sequence models with attention, Connectionist Temporal Classification and the RNN Sequence Transducer are currently supported.

The goal of this software is to facilitate research in end-to-end models for speech recognition. The models are implemented in PyTorch.

The software has only been tested in Python3.6.

We will not be providing backward compatability for Python2.7.

We recommend creating a virtual environment and installing the python requirements there.

virtualenv <path_to_your_env>
source <path_to_your_env>/bin/activate
pip install -r requirements.txt

Then follow the installation instructions for a version of PyTorch which works for your machine.

After all the python requirements are installed, from the top level directory, run:

make

The build process requires CMake as well as Make.

After that, source the setup.sh from the repo root.

source setup.sh

Consider adding this to your bashrc.

You can verify the install was successful by running the tests from the tests directory.

cd tests
pytest

To train a model run

python train.py <path_to_config>

After the model is done training you can evaluate it with

python eval.py <path_to_model> <path_to_data_json>

To see the available options for each script use -h:

python {train, eval}.py -h

For examples of model configurations and datasets, visit the examples directory. Each example dataset should have instructions and/or scripts for downloading and preparing the data. There should also be one or more model configurations available. The results for each configuration will documented in each examples corresponding README.md.