Keyword Spotting Transformer

This is the unofficial TensorFlow implementation of the Keyword Spotting Transformer model. This model is used to train on the 35 words speech command dataset

Paper : Keyword Transformer: A Self-Attention Model for Keyword Spotting

Model architecture

alt text

Download the dataset

To download the dataset use the following command

mkdir data
mv ./speech_commands_v0.02.tar.gz ./data
cd ./data
tar -xf ./speech_commands_v0.02.tar.gz
cd ../

Setup virtual environment

virtualenv -p python3 venv
source ./venv/bin/activate

Install dependencies

pip install -r requirements.txt

Training the model

To train the model run this command

python3 --data_dir ${Path to data directory} \
                 --logdir ${Path to log directory} \
                 --num_layers ${Number of sequential encoder layers} \
                 --d_model ${Dimension of the encoder layers} \
                 --num_heads ${Number of heads in multi head attention layer} \
                 --mlp_dim ${Dimension of mlp layers} \
                 --lr ${Learning rate} \
                 --weight_decay ${Weight decay} \
                 --batch_size ${Batch size} \
                 --epochs ${Number of epochs} \
                 --save_dir ${Directory to save the model weights}

To track your training metrics

tensorboard --logdir  ${Path to log directory}

Predicting keyword of audio file

To predict the keyword of the audio file

python3 --model_dir ${Saved model directory} \
                --file_path ${Audio file}