wekws/runtime/android

Usage

Most of AI engineers are not familiar with Android development, this is a simple ‘how to’.

1. Train your model with your data

2. Export pytorch model to onnx model

3. Convert onnx model for mobile deployment

python -m onnxruntime.tools.convert_onnx_models_to_ort your-model.onnx

you will get your-model.ort and your-model.with_runtime_opt.ort

4. Install Android Studio and open path of wekws/runtime/android and build

NOTE: The default feature_dim in code is 40, if your model’s is 80, change it here ./app/src/main/cpp/wekws.cc

  feature_config = std::make_shared<wenet::FeaturePipelineConfig>(40, 16000);  // 40 -> 80

It’s also can be built on Linux by runing bash ./gradlew build

5. Install app/build/outputs/apk/debug/app-debug.apk to your phone and try it.