treble-technologies/Treble10-Speech

Treble10-Speech 是一个自动语音识别（ASR）数据集，包含使用高保真房间声学模拟的预卷积语音文件。数据集包含 10 个不同的家具房间，包括 2 个浴室、2 个卧室、2 个带走廊的客厅、2 个不带走廊的客厅和 2 个会议室。房间体积在 14 到 46 立方米之间，混响时间在 0.17 到 0.84 秒之间。数据集分为三个子集：Treble10-Speech-mono、Treble10-Speech-hoa8 和 Treble10-Speech-6ch，分别对应单声道、8 阶 Ambisonics 编码和 6 通道设备的混响语音文件。数据集可用于远场 ASR、语音增强、去混响和源分离等任务。

The Treble10-Speech dataset is an automatic speech recognition (ASR) dataset containing pre-convolved speech files using high fidelity room-acoustic simulations from the Treble10-RIR dataset with 10 different furnished rooms. The dataset is divided into three subsets: Treble10-Speech-mono, Treble10-Speech-hoa8, and Treble10-Speech-6ch, corresponding to mono, 8th-order Ambisonics encoded, and 6-channel device microphone speech files, respectively. The dataset is suitable for tasks such as far-field ASR, speech enhancement, dereverberation, and source separation.