Summary

  • 3 minutes

Congratulations on learning how to build a binary audio classification model.

You learned how analog sound is represented as digital samples, how waveforms show amplitude over time, and how spectrograms show frequency content over time. You also saw how to load the yes and no classes from the Speech Commands dataset, convert waveforms into spectrogram tensors, train a convolutional neural network, and evaluate the model on held-out data.

Next steps

To go deeper, try these ideas:

  • Extend the model to classify all eight commands in the mini Speech Commands dataset by loading SOURCE_DATASET_PATH instead of the binary BINARY_DATASET_PATH. The label_names array and num_labels value update automatically from the directory names.
  • Use the full Speech Commands dataset and the train, validation, and test splits provided by TensorFlow Datasets for a more rigorous evaluation. TensorFlow Datasets exposes the standard keyword-spotting label set, including target commands plus _unknown_ and _silence_, so update the output layer and label handling for that label set.
  • Try transfer learning for audio recognition with a pretrained YAMNet model.