ExecuTorch + KissFFT SPEECHCOMMANDS C++ Application

This is a simple C++ demo application that uses the ExecuTorch library and KissFFT for SPEECHCOMMANDS CNN model inference.

Build instructions

Export the model (model.ipynb) or use a PTE model already provided.
The ExecuTorch repository and KissFFT repository need to be downloaded to third-party directory.
```
 mkdir third-party
 cd third-party
 git clone https://github.com/pytorch/executorch.git
 git clone https://github.com/mborgerding/kissfft.git
```
Tested versions: 39c9781 for kissfft and 8c84780 for executorch.
Install dev requirements for ExecuTorch (venv advised):
```
pip install -r requirements.txt
```
Build the project (python source must have executorch installed):
```
./build.sh
```

Run the demo application:

./build/bin/speechcommands_app cat_test.wav

Make sure you have the correct model file (.pte) compatible with ExecuTorch.
This app can be tested with wav files (from SPEECHCOMMANDS etc.) - it uses first second of the recording (must be sorted wav) to infer the word.