A Pytorch Implementation of Transducer Model for End-to-End Speech Recognition.
Email: zhengkun.tian@nlpr.ia.ac.cn
- pytorch >= 0.4
- warp-transducer
We utilize Kaldi for data preparation. At least these files(text, feats.scp) should be included in the training/development/test set. If you apply cmvn, utt2spk and cmvn.scp are required. The format of these file is consistent with Kaidi. The format of vocab is as follows.
<blk> 0
<unk> 1
我 2
你 3
...
python train.py -config config/aishell.yaml
python eval.py -config config/aishell.yaml
The details of our RNN-Transducer are as follows.
model:
enc:
type: lstm
hidden_size: 320
n_layers: 4
bidirectional: True
dec:
type: lstm
hidden_size: 512
n_layers: 1
embedding_dim: 512
vocab_size: 4232
dropout: 0.2
All experiments are conducted on AISHELL-1. During decoding, we use beam search with width of 5 for all the experiments. A character-level 5-gram language model from training text, is integrated into beam searching by shallow fusion.
MODEL | DEV(CER) | TEST(CER) |
---|---|---|
RNNT+pretrain+LM | 10.13 | 11.82 |
Thanks to warp-transducer.