Efficient Conformer V2

cd ../../../../toolbox/WeNet/
git clone https://github.com/wenet-e2e/wenet.git
cd wenet/
sed -i 's/^torch/# torch/g' requirements.txt
pip3 install -r requirements.txt

# Change to the scripts path
cd wenet/examples/aishell/s0/

# Add $data_path and $model_name to run.sh
sed -i s/^data=.*/data=\${data_path}/g run.sh
sed -i s#^train_config=.*#train_config=conf/train_\${model_name}#g run.sh
sed -i s#^dir=.*#dir=exp/\${model_name}#g run.sh

# Configure data path and model name
export data_path="/path/to/aishell"
export model_name="u2++_efficonformer_v2"

# Add torchrun command to PATH
ln -s /usr/local/corex-3.1.0/lib64/python3/dist-packages/bin/torchrun /usr/local/bin/

# Run all stages
bash run.sh --stage -1 --stop-stage 6

# Download data
bash run.sh --stage -1 --stop-stage -1
# Prepare Training data
bash run.sh --stage 0 --stop-stage 0
# Extract optinal cmvn features
bash run.sh --stage 1 --stop-stage 1
# Generate label token dictionary
bash run.sh --stage 2 --stop-stage 2
# Prepare WeNet data format
bash run.sh --stage 3 --stop-stage 3
# Neural Network training
bash run.sh --stage 4 --stop-stage 4
# Recognize wav using the trained model
bash run.sh --stage 5 --stop-stage 5
# Export the trained model
bash run.sh --stage 6 --stop-stage 6