CROWDSPEECH, VOXDIY

CROWDSPEECH是首个公开的大型众包音频转录数据集，由Yandex和Carnegie Mellon University的研究人员共同创建。该数据集包含超过60小时的英语语音，由3994名众包工作者完成转录。此外，VOXDIY是CROWDSPEECH的俄语版本，用于在资源较少的领域中构建众包音频转录数据集。这些数据集的创建旨在解决机器学习系统从基准到实际应用的转移问题，特别是在语音识别领域。通过这些数据集，研究人员可以评估和改进现有的聚合方法，并设计新的算法以提高众包语音注释的质量。

CROWDSPEECH is the first publicly available large-scale crowdsourced audio transcription dataset, jointly created by researchers from Yandex and Carnegie Mellon University. This dataset includes over 60 hours of English speech, with transcriptions completed by 3994 crowdworkers. Additionally, VOXDIY is the Russian-language adaptation of CROWDSPEECH, developed for building crowdsourced audio transcription datasets in low-resource domains. The creation of these datasets aims to solve the domain shift problem faced when transitioning machine learning systems from benchmark evaluations to real-world applications, particularly in the field of speech recognition. With these datasets, researchers can evaluate and improve existing aggregation methods, as well as design novel algorithms to enhance the quality of crowdsourced speech annotations.