model_convert.md

模型转换

运行环境

• Clone the repo

git clone https://github.com/junruizh2021/keyword-spotting.git && cd keyword-spotting

• Install Conda: please see https://docs.conda.io/en/latest/miniconda.html

• Install miniconda

conda create -n kws python=3.9
conda activate kws
pip install -r requirements.txt

下面示例模型转换，示例均使用“你好问问”数据训练的模型。

Max-pooling方案模型转换

1.下载模型

git clone https://www.modelscope.cn/daydream-factory/keyword-spot-dstcn-maxpooling-wenwen.git

模型目录结构如下 >>

$ tree keyword-spot-dstcn-maxpooling-wenwen

keyword-spot-dstcn-maxpooling-wenwen
├── avg_30.pt
├── configuration.json
├── config.yaml
├── global_cmvn
├── README.md

2.模型转换

2.1 pytorch to onnx

python model_convert/export_onnx.py \
 --config keyword-spot-dstcn-maxpooling-wenwen/config.yaml \
 --checkpoint keyword-spot-dstcn-maxpooling-wenwen/avg_30.pt \
 --onnx_model keyword-spot-dstcn-maxpooling-wenwen/onnx/keyword-spot-dstcn-maxpooling-wenwen.onnx

2.2 onnx2ort. 用于端侧设备部署.

python -m onnxruntime.tools.convert_onnx_models_to_ort keyword-spot-dstcn-maxpooling-wenwen/onnx/keyword-spot-dstcn-maxpooling-wenwen.onnx

输出模型结构

tree keyword-spot-dstcn-maxpooling-wenwen

keyword-spot-dstcn-maxpooling-wenwen
├── avg_30.pt
├── configuration.json
├── config.yaml
├── global_cmvn
├── onnx
│   ├── keyword-spot-dstcn-maxpooling-wenwen.onnx  #中间模型
│   ├── keyword-spot-dstcn-maxpooling-wenwen.ort   #用于端侧部署的ort模型
│   ├── keyword-spot-dstcn-maxpooling-wenwen.required_operators.config
│   ├── keyword-spot-dstcn-maxpooling-wenwen.required_operators.with_runtime_opt.config
│   └── keyword-spot-dstcn-maxpooling-wenwen.with_runtime_opt.ort
├── README.md
└── words.txt

CTC方案模型转换

[To-do]