Athena-Decoder is an open-source Automatic Speech Recognition (ASR) decoder tool implemented in python.
Our vision is to offer an universal, standard and user-friendly decoder tool for all kinds of model including but not limited to CTC, Attention-based model and Transformer
This project has only been tested on Python 3. We recommend creating a virtual environment and installing the python requirements there.
git clone https://github.com/athena-team/athena-decoder.git
pip install -r requirements.txt
python setup.py bdist_wheel sdist
python -m pip install --ignore-installed dist/pydecoders-0.1.0*.whl
source ./tools/env.sh
python pydecoders/build_graph_main.py examples/hkust/conf/hkust.conf
# prepare data
cd examples/aishell && ./prepare_data.sh
# build graph
python pydecoders/build_graph_main.py examples/aishell/conf/aishell.conf
# first build WFST graph,optional
# or you can skip this build step and use graph in hkust/graph directly
python pydecoders/build_graph_main.py
# run decode step
python pydecoders/decode_wfst_main.py examples/hkust/conf/hkust.conf
# run beam search decoder
python pydecoders/decode_beam_search_main.py examples/hkust/conf/hkust.conf
Please reference to build-graph.md and decoders.md in docs directory in detail.