Libri-adhoc40

Libri-adhoc40数据集是由中国西北工业大学海洋科学与技术学院创建，专注于同步的临时麦克风阵列研究。该数据集包含4510小时的数据，每麦克风110小时，源自Librispeech的‘train-clean-100’, ‘dev-clean’和‘test-clean’子集。数据收集过程涉及在真实办公室环境和消声室内通过40个强同步分布节点重放Librispeech数据。该数据集旨在解决远场语音处理问题，支持语音前端处理等多种应用，为多设备语音识别系统提供基准测试。

The Libri-adhoc40 dataset was developed by the School of Marine Science and Technology, Northwestern Polytechnical University, China, and focuses on research into synchronous ad-hoc microphone arrays. This dataset contains 4510 hours of audio data, with 110 hours per microphone, sourced from the 'train-clean-100', 'dev-clean', and 'test-clean' subsets of the Librispeech corpus. The data collection procedure entails replaying Librispeech audio content through 40 tightly synchronized distributed nodes in both real-world office environments and anechoic chambers. This dataset is designed to address far-field speech processing challenges, supports a wide range of applications including speech front-end processing, and serves as a benchmark for multi-device speech recognition systems.