Skip to content

Audio classification with VGGish as feature extractor in TensorFlow

License

Notifications You must be signed in to change notification settings

luuil/Tensorflow-Audio-Classification

Repository files navigation

Audio Classification

Classify the audios. In this repo, I train a model on UrbanSound8K dataset, and achieve about 80% accuracy on test dataset.

There is a pre-trained model in urban_sound_train, trained epoch is 1000

Usage

Env setup

Conda are recommended, just need one line: conda env create -f conda.env.yml

Train & Test

  1. Config parameters: audio_params.py.
  2. Train the model: python audio_train.py. (It will create tfrecords automaticly if not exists)
  3. Check the training process from tensorboard: tensorboard --logdir=./data/tensorboard
  4. Test the model: python audio_inference_demo.py.

Tools

Dataset

Ref. Blogs