Warden Dataset

该数据集包含了105,829个一秒钟长的音频剪辑，涵盖了35个常见词汇，其中包括对物联网和机器人技术有用的数字和词汇，以及涵盖多种音素的额外词汇。数据集中既有目标词汇的正样本，也有背景噪音或不相关语音的负样本，这有助于提升关键词识别系统的性能。此外，该数据集还用于评估音频增强方法的效果，尤其是在区分相关与非相关音频剪辑方面的表现。该数据集的规模为105,829个剪辑，任务是对语音命令的识别（Scr）。

This dataset consists of 105,829 one-second audio clips covering 35 common words, including numerals and terms useful for the Internet of Things (IoT) and robotics, as well as additional words spanning a variety of phonemes. The dataset includes both positive samples of target words and negative samples of background noise or irrelevant speech, which helps improve the performance of keyword recognition systems. Furthermore, this dataset can be used to evaluate the effectiveness of audio enhancement methods, especially their performance in distinguishing relevant from irrelevant audio clips. The dataset has 105,829 clips in total, and the corresponding task is speech command recognition (Scr).