QCRI/DACS

DACS数据集是一个包含2小时标注埃及阿拉伯语语音的数据集，这些语音来自于ADI-5开发分割的MGB-3挑战。该数据集已根据500ms以上的静默手动切分为更小的发音单元，并由埃及本地非专业人士进行了逐字转录。数据集同时提供了自动语音识别转录的语音特征和文本特征，并进行了词级别的代码切换标注。The DACS dataset includes 2 hours of annotated Egyptian Arabic speech from the ADI-5 development split in the MGB-3 Challenge. The corpus has been manually segmented into smaller utterances based on pauses of ≥500 ms of silence and transcribed verbatim by a lay native Egyptian speaker. The dataset provides both speech features and textual features extracted from ASR transcriptions, and it has been annotated for word-level Code-Switching by three independent annotators.