环境搭建
新建虚拟环境:conda create -n tensorflow python=3.6
激活虚拟环境:source activate tensorflow
1.安装tensorflow:pip install -i https://pypi.tuna.tsinghua.edu.cn/simple tensorflow==1.13.1
2.安装keras:pip install -i https://pypi.tuna.tsinghua.edu.cn/simple keras==2.2.4
3.安装语音流处理模块:pip install -i https://pypi.tuna.tsinghua.edu.cn/simple soundfile==0.10.2
训练环境安装前三个就可以,测试环境需要后面两个
4.安装beam search解码模块(解码模块使用mozilla项目里面的):pip install https://index.taskcluster.net/v1/task/project.deepspeech.deepspeech.native_client.v0.5.0-alpha.11.cpu-ctc/artifacts/public/ds_ctcdecoder-0.5.0a11-cp36-cp36m-manylinux1_x86_64.whl
报错platform不支持的话在mozilla的DeepSpeech里面执行进行安装:pip install $(python util/taskcluster.py --decoder)
gpu版:pip install https://index.taskcluster.net/v1/task/project.deepspeech.deepspeech.native_client.v0.5.0-alpha.11.cpu-ctc/artifacts/public/ds_ctcdecoder-0.5.0a11-cp35-cp35m-manylinux1_x86_64.whl
5.读字节流:pip install -i https://pypi.tuna.tsinghua.edu.cn/simple pydub