audio_classifier

Audio classification using Tensorflow Lite

This example uses a simple CNN model that recognizes a keyword from an audio file. to showcase the working of TFLite delegate running with ArmNN acceleration.

Refer to the audio_classifier_model.ipynb notebook on how to build this model yourself.

Requirements for running the notebook:

tensorflow==2.15.0
matplotlib==3.8.2
librosa==0.10.1

You will need to have the following file structure for the dataset.

├── audio_classifier_model.ipynb
└── datasets
    ├── edge
    ├── hello
    ├── khadas
    ├── mind
    ├── none
    ├── tone
    └── vim

Each subfolder should have 20, 5 second duration .wav files named in order from 0 to 19, such as edge-0.wav, edge-1.wav,... edge-19.wav

This model can identify the following keywords 0. none (no keyword said)

1. hello
2. khadas
3. vim
4. edge
5. tone
6. mind

Add symlinks for libraries

sudo ln ../libs/delegate/libarmnnDelegate.so.29.1 libarmnnDelegate.so.29
sudo ln ../libs/libarmnn.so.34.0 libarmnn.so.34

Run the example using sample.wav audio

python3 run_inference.py

Run the example using microphone audio

python3 run_inference.py m

Note:

Modify the BACKEND variable in the code to use either GpuAcc or CpuAcc backends.