icefall/egs/mgb2/ASR/RESULTS.md

Results

MGB2 BPE training results (Conformer-CTC) (after 3 epochs)

2022-06-04

The best WER, as of 2022-06-04, for the MGB2 test dataset is below

Using whole lattice HLG decoding + n-gram LM rescoring + attention decoder rescoring

	dev	test
WER	25.32	23.53

Scale values used in n-gram LM rescoring and attention rescoring for the best WERs are:

ngram_lm_scale	attention_scale
0.1	-

Using n-best (n=0.5) HLG decoding + n-gram LM rescoring + attention decoder rescoring:

	dev	test
WER	27.87	26.12

Scale values used in n-gram LM rescoring and attention rescoring for the best WERs are:

ngram_lm_scale	attention_scale
0.01	0.3

To reproduce the above result, use the following commands for training:

Note: the model was trained on V-100 32GB GPU

cd egs/mgb2/ASR
. ./path.sh
./prepare.sh
export CUDA_VISIBLE_DEVICES="0,1"
./conformer_ctc/train.py \
  --lang-dir data/lang_bpe_5000 \
  --att-rate 0.8 \
  --lr-factor 10 \
  --max-duration  \
  --concatenate-cuts 0 \
  --world-size 2 \
  --bucketing-sampler 1 \
  --max-duration 100 \ 
  --start-epoch 0 \
  --num-epochs 30
  

and the following command for nbest decoding

./conformer_ctc/decode.py \
  --lang-dir data/lang_bpe_5000 \
  --max-duration 30 \
  --concatenate-cuts 0 \
  --bucketing-sampler 1 \
  --num-paths 1000 \
  --epoch 2 \
  --avg 2 \
  --method attention-decoder \
  --nbest-scale 0.5

and the following command for whole-lattice decoding

./conformer_ctc/decode.py \
  --lang-dir data/lang_bpe_5000 \
  --max-duration 30 \
  --concatenate-cuts 0 \
  --bucketing-sampler 1 \
  --num-paths 1000 \
  --epoch 2 \
  --avg 2 \
  --method  whole-lattice-rescoring

You can find the pre-trained model by visiting

The tensorboard log for training is available at https://tensorboard.dev/experiment/zy6FnumCQlmiO7BPsdCmEg/#scalars