luomingshuang's picture
add README.d
0e7bfd0

Note: This recipe is trained with the codes from this PR https://github.com/k2-fsa/icefall/pull/355 And the SpecAugment codes from this PR https://github.com/lhotse-speech/lhotse/pull/604.

Pre-trained Transducer-Stateless2 models for the Aidatatang_200zh dataset with icefall.

The model was trained on full Aidatatang_200zh with the scripts in icefall based on the latest version k2.

Training procedure

The main repositories are list below, we will update the training and decoding scripts with the update of version.
k2: https://github.com/k2-fsa/k2 icefall: https://github.com/k2-fsa/icefall lhotse: https://github.com/lhotse-speech/lhotse

git clone https://github.com/k2-fsa/icefall
cd icefall
  • Preparing data.
cd egs/aidatatang_200zh/ASR
bash ./prepare.sh
  • Training
export CUDA_VISIBLE_DEVICES="0,1"
./pruned_transducer_stateless2/train.py \
                  --world-size 2 \
                  --num-epochs 30 \
                  --start-epoch 0 \
                  --exp-dir pruned_transducer_stateless2/exp \
                  --lang-dir data/lang_char \
                  --max-duration 250

Evaluation results

The decoding results (WER%) on Aidatatang_200zh(dev and test) are listed below, we got this result by averaging models from epoch 11 to 29. The WERs are

dev test comment
greedy search 5.53 6.59 --epoch 29, --avg 19, --max-duration 100
modified beam search (beam size 4) 5.28 6.32 --epoch 29, --avg 19, --max-duration 100
fast beam search (set as default) 5.29 6.33 --epoch 29, --avg 19, --max-duration 1500