0x3/DNS-Challenge

该数据集是INTERSPEECH 2020深度噪声抑制挑战（DNS Challenge）的一部分，用于训练和评估语音增强模型。数据集包含干净语音和噪声片段，其中干净语音来自多个公开来源，如LibriVox（公共领域）、PTDB-TUG（Graz University of Technology的语音数据库）和爱丁堡大学56说话者数据集；噪声数据来自Audioset（Creative Commons许可）、Freesound（仅CC0许可文件）和Demand数据库。数据集通过提供的脚本（如noisyspeech_synthesizer_singleprocess.py）可合成带噪-干净语音对，支持深度噪声抑制任务的研究和开发。This dataset is part of the INTERSPEECH 2020 Deep Noise Suppression Challenge (DNS Challenge), designed for training and evaluating speech enhancement models. It includes clean speech and noise clips, with clean speech sourced from multiple public repositories such as LibriVox (public domain), PTDB-TUG (Pitch Tracking Database from Graz University of Technology), and the Edinburgh 56 speaker dataset; noise data is derived from Audioset (Creative Commons licensed), Freesound (only CC0 licensed files), and the Demand database. The dataset can be synthesized using provided scripts (e.g., noisyspeech_synthesizer_singleprocess.py) to generate noisy-clean speech pairs, facilitating research and development in deep noise suppression tasks.