Title: A long-form single-speaker real-time MRI speech dataset and benchmark Show Chinese title

Title: 一种长格式单说话人实时MRI语音数据集和基准

Abstract: We release the USC Long Single-Speaker (LSS) dataset containing real-time MRI video of the vocal tract dynamics and simultaneous audio obtained during speech production. This unique dataset contains roughly one hour of video and audio data from a single native speaker of American English, making it one of the longer publicly available single-speaker datasets of real-time MRI speech data. Along with the articulatory and acoustic raw data, we release derived representations of the data that are suitable for a range of downstream tasks. This includes video cropped to the vocal tract region, sentence-level splits of the data, restored and denoised audio, and regions-of-interest timeseries. We also benchmark this dataset on articulatory synthesis and phoneme recognition tasks, providing baseline performance for these tasks on this dataset which future research can aim to improve upon. Show Chinese abstract

Abstract: 我们发布了USC长单说话人（LSS）数据集，其中包含语音产生过程中实时MRI视频和同时获得的音频。 这个独特的数据集包含一位美式英语母语者大约一小时的视频和音频数据，使其成为公开可用的较长的实时MRI语音数据单说话人数据集之一。 除了运动和声学原始数据外，我们还发布了适合多种下游任务的数据衍生表示。 这包括裁剪到声道区域的视频、数据的句子级划分、恢复和去噪的音频以及感兴趣区域的时间序列。 我们还在运动合成和音素识别任务上对该数据集进行了基准测试，为这些任务提供了该数据集的基线性能，未来的研究可以以此为目标进行改进。