Speech Commands Classification is a project that aims to classify the speech commands from the Speech Commands Dataset using a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) network. In this project we will use PyTorch to build the CNN to classify the speech commands.

My recommendation is to install with requirements.txt file.

pip install -r requirements.txt

Some of you probably have different cuda or gpu, so you can install PyTorch with this tool.

You can download the new version from here. Extract the dataset in the directory you wish to work in. Split the dataset into train, validation and test sets using the following command:

python main.py --action create_dataset --speech_commands_folder <google-command-folder> --out_path <path to save the splitted data>

You can leave speech_commands_folder and out_path empty to use the default values.

To Transform the dataset from wav files to numpy array files you can use this command:

python main.py --action transform_dataset --dataset <path-to-dataset-folder>

You can leave dataset empty to use the default values.

The architectures of the CNN is the following:

• Standard LeNet.
• LeNet with several modifications. Which includes:
  • Batch normalization.
  • Dropout
  • ReLU activation function.
• AlexNet.

If you want to train the CNN you can use the following command:

python main.py --action train --dataset <path-to-dataset-folder> --model_name <'lenet', 'improved_lenet', 'alexnet'>

There are several parameters you can use to train the CNN. You can see them by using the following command:

python main.py --help

For testing the CNN you can use the following command:

python main.py --action test --dataset <path-to-dataset> --model_checkpoint <path-to-model>

You can see the validation accuracy and F1 for each epoch in the following figures:

Lenet in gray. Improved LeNet in blue. AlexNet in purple.