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

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

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

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

Its also can be built on Linux by runing bash ./gradlew build

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