Jason-1998/RNN-Transducer

MXNet implementation of RNN Transducer (Graves 2012): Sequence Transduction with Recurrent Neural Networks

RNN Transducer

MXNET GPU version of RNN Transducer loss is now available !

File description

• eval.py: transducer decode
• model.py: rnn transducer refer to Graves2012
• DataLoader.py: data process
• train.py: rnnt training script, can be initialized from CTC and PM model

Directory description

• conf: kaldi feature extraction config

Reference Paper

• RNN Transducer (Graves 2012): Sequence Transduction with Recurrent Neural Networks
• RNNT joint (Graves 2013): Speech Recognition with Deep Recurrent Neural Networks

Run

• Compile RNNT Loss Follow the instructions in here to compile MXNET with RNNT loss.

• Extract feature link kaldi timit example dirs (local steps utils ) excute run.sh to extract 13 dim mfcc feature run feature_transform.sh to get 26 dim feature as described in Graves2012

• Train RNNT model:

python train.py --lr 1e-3 --bi --dropout .5 --out exp/rnnt_bi_lr1e-3 --schedule

Evaluation

Default only for RNNT

• Greedy decoding:

python eval.py <path to best model parameters> --bi

• Beam search:

python eval.py <path to best model parameters> --bi --beam <beam size>

Results

• CTC

  Decode	PER
  greedy	22.27
  beam 20	21.83

• Transducer

  Decode	PER
  greedy	23.02
  beam 20	22.45
  beam 40	22.34
  beam 60	21.92
  beam 80	22.15

TODO

• beam serach accelaration
• several baseline
• Seq2Seq with attention