This project has been fully tested on Python 3.6.8
and TensorFlow 2.0.0
under Ubuntu 18.04.2 LTS
.
We recommend that users use Docker
or a virtual environment such as conda
to install the python requirements.
conda create -p <path>/<env_name> python=3.6
source activate <path>/<env_name>
conda install tensorflow-gpu=2.0.0
Delta dependient on third party tools, so when run the program, need blow to install tools:
activate the environment and use below
cd tools && make
For case you want install Tensorflow Gpu 2.0.0
, under machine which has Gpu Driver 410.48
.
It has problem of runtime not compariable with driver version, when isntall using conda.
Then we can install tensorflow from Pip
as below:
Same to conda install.
See CUDA Compatibility for CUDA Toolkit and Compatible Driver Version
.
See cuDNN Support Matrix for cuDNN For CUDA and NVIDIA Hardware
.
For Nvidia Driver Version: 418.67, CUDA Version: 10.1:
conda install cudatoolkit==10.1.168-0
conda install cupti=10.1.168-0
conda install cudnn==7.6.0
or
conda install cudatoolkit==10.1
conda install cupti==10.1
conda install cudnn==7.6.0
For user in China, we can set conda mirror as below:
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/
conda config --set show_channel_urls yes
Other references: conda-forge tuna
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple tensorflow-gpu==2.0.0
For tensorflow 2.0.0, make sure numpy version is 1.16.4.
Same to conda install.
Install DELTA without speech
dependences:
cd tools && make basic check_install
By default we will install DELTA with Kaldi
toolkit:
cd tools && make delta
If user has installed Kaldi
, please DELTA as below:
cd tools && make delta KALDI=<kaldi-path>
it is simply link the <kaldi-path>
to tools/kaldi
.
Please see delta
target of tools/Makefile
.
Install DELTANN as below:
cd tools && make deltann
For more details, please see deltann
target of tools/Makefile