CiwaGAN Fine-tuning Dataset
收藏arXiv2025-09-30 收录
下载链接：
http://doi.org/10.17605/OSF.IO/JBWYH
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了大约1700个标记，用于对Whisper ASR模型进行微调，它融合了TIMIT词汇项和CiwaGAN模型手动标注的输出。具体来说，数据集包括了TIMIT数据集中的100个标记，这些标记对应于训练中使用的9个单词，同时结合了作者标注的800个样本。规模上，该数据集大约有1700个标记，其任务是进行语音识别的微调。
This dataset contains approximately 1700 labeled instances for fine-tuning Whisper ASR models, and it combines TIMIT lexical items with manually annotated outputs from the CiwaGAN model. Specifically, the dataset includes 100 labeled instances from the TIMIT corpus, which correspond to the 9 words utilized in training, along with 800 samples annotated by the authors. In terms of scale, this dataset has roughly 1700 labeled instances, and its core task is speech recognition fine-tuning.
提供机构：
Authors of the paper
