ESPNET 安装指引

准备Python环境

  1. 建议安装anaconda3: anaconda3
  2. 创建ESPNET的环境: conda create -n espnet python=3.6

Install ESPNET相关依赖

  1. 下载espnet源代码 espnet github
  2. pip install torch==1.0.1 # Installation from anaconda is recommended for PyTorch
  3. pip install chainer==6.0.0
  4. pip install cupy==6.0.0 # Do not install cupy as default
  5. pip install setuptools>=38.5.1
  6. pip install scipy>=1.3.0
  7. pip install librosa>=0.7.0
  8. pip install soundfile>=0.10.2
  9. pip install inflect>=1.0.0
  10. pip install unidecode>=1.0.22
  11. pip install editdistance==0.5.2
  12. pip install h5py==2.9.0
  13. pip install tensorboardX==1.8
  14. pip install pillow>=6.1.0
  15. pip install nara_wpe>=0.0.5
  16. pip install museval>=0.2.1
  17. pip install pystoi>=0.2.2
  18. pip install kaldiio>=2.13.8
  19. pip install matplotlib>=3.1.0
  20. pip install funcsigs>=1.0.2 # A backport of inspect.signature for python2
  21. pip install configargparse>=0.14.0
  22. pip install PyYAML>=5.1.2
  23. pip install sentencepiece>=0.1.82
  24. pip install pysptk>=0.1.17
  25. pip install nltk>=3.4.5
  26. pip install nnmnkwii
  27. pip install jaconv
  28. pip install g2p_en
  29. pip install torch_complex@git+https://github.com/kamo-naoyuki/pytorch_complex.git
  30. pip install pytorch_wpe@git+https://github.com/nttcslab-sp/dnn_wpe.git
  31. pip install warpctc-pytorch10-cuda90==0.1.3 (GitHub源码不易安装,pip安装较为简单)

Install 其他依赖

  1. kaldi相关的依赖: sudo yum install atlas
  2. sudo yum install bc
  3. sudo yum install tree
  4. sudo yum install libsndfile
  5. sudo yum install flac (librispeech 语音转格式,需要安装flac)

install cuda

  1. 目前已支持cuda10,安装 sh /path/to/cuda10/cuda_10.0.130_410.48_linux.run
  2. export LD_LIBRARY_PATH=/path/to/cuda/lib64:$LD_LIBRARY_PATH

Install ESPNET

  1. conda activate espnet (或者recipe里面的 path.sh)
  2. pip install -e /path/to/espnet/

Run ESPNET

  1. conda activate espnet
  2. cd /path/to/espnet/egs/recipe
  3. sh run.sh