Speech Commands: A Dataset for Limited-Vocabulary Speech Recognition

TL;DR

This paper introduces a specialized audio dataset for limited-vocabulary speech recognition, facilitating the development and evaluation of keyword spotting systems with reproducible metrics.

Contribution

It provides a new, well-structured dataset tailored for keyword spotting, along with a methodology for consistent evaluation and baseline results.

Findings

Dataset enables effective training of keyword spotting models

Reproducible accuracy metrics established for the task

Baseline model results demonstrate dataset utility

Abstract

Describes an audio dataset of spoken words designed to help train and evaluate keyword spotting systems. Discusses why this task is an interesting challenge, and why it requires a specialized dataset that is different from conventional datasets used for automatic speech recognition of full sentences. Suggests a methodology for reproducible and comparable accuracy metrics for this task. Describes how the data was collected and verified, what it contains, previous versions and properties. Concludes by reporting baseline results of models trained on this dataset.