pykaldi2
PyKaldi2 is a speech toolkit that is built based on Kaldi and PyTorch. It relies on PyKaldi -- the Python wrapper of Kaldi, to access Kaldi functionalities. The key features of PyKaldi2 are one-the-fly lattice generation for lattice-based sequence training, on-the-fly data simulation and on-the-fly alignment gereation.
How to install
PyKaldi2 runs on top of Horovod and PyKaldi libraries. The dockerfile is provided to customarize the envriorment. To use the repo, do the following three steps.
-
Clone the repo by
git clone https://github.com/jzlianglu/pykaldi2.git -
Build the docker image, simply run
docker build -t horovod-pykaldi -f docker/dockerfile
- Activate the docker image, for example
NV_GPU=0,1,2,3 nvidia-docker run -v `pwd`:`pwd` -w `pwd` --shm-size=32G -i -t horovod-pykaldi
If you want to run multi-GPU jobs using Horovod, command is like
horovodrun -np 4 -H localhost:4 sh run_ce.sh
Cross-entropy training
An example of runing a cross-entropy job is
python train_ce.py -config configs/ce.yaml \
-data configs/data.yaml \
-exp_dir exp/tr960_blstm_3x512_dp02 \
-lr 0.0002 \
-batch_size 64 \
-sweep_size 960 \
-aneal_lr_epoch 3 \
-num_epochs 8 \
-aneal_lr_ratio 0.5
Sequence training
An example of runing a sequence traing job is
python train_se.py -config configs/mmi.yaml \
-data configs/data.yaml \
-exp_dir exp/tr960_blstm_3x512_dp02 \
-criterion "mmi" \
-seed_model exp/tr960_blstm_3x512_dp02/model.7.tar \
-prior_path /datadisk2/lial/librispeech/s5/exp/tri6b_ali_tr960/final.occs \
-trans_model /datadisk2/lial/librispeech/s5/exp/tri6b_ali_tr960/final.mdl \
-den_dir /datadisk2/lial/librispeech/s5/exp/tri6b_denlats_960_cleaned/dengraph \
-lr 0.000001 \
-ce_ratio 0.1 \
-max_grad_norm 5 \
-batch_size 4 \
Reference
Liang Lu, Xiong Xiao, Zhuo Chen, Yifan Gong, "PyKaldi2: Yet another speech toolkit based on Kaldi and PyTorch", arxiv, 2019