Honk: A PyTorch Reimplementation of Convolutional Neural Networks for Keyword Spotting

TL;DR

Honk is an open-source PyTorch implementation of CNN models for keyword spotting, providing a comparable accuracy baseline on Google's Speech Commands Dataset to facilitate future research in speech command recognition.

Contribution

It offers a PyTorch reimplementation of CNN-based keyword spotting models originally in TensorFlow, aiding reproducibility and further development.

Findings

Comparable accuracy to original TensorFlow models

Provides a practical PyTorch baseline for keyword spotting

Facilitates future research in speech command recognition

Abstract

We describe Honk, an open-source PyTorch reimplementation of convolutional neural networks for keyword spotting that are included as examples in TensorFlow. These models are useful for recognizing "command triggers" in speech-based interfaces (e.g., "Hey Siri"), which serve as explicit cues for audio recordings of utterances that are sent to the cloud for full speech recognition. Evaluation on Google's recently released Speech Commands Dataset shows that our reimplementation is comparable in accuracy and provides a starting point for future work on the keyword spotting task.