Interspeech 2025 URGENT Challenge Track 1

该数据集包含了用于训练和验证语音增强模型的噪声-清晰数据对，涵盖了多种失真情况，如数据包丢失、噪声、混响和剪辑等。官方的盲测集包括50%的真实录音和50%的合成数据，同时还包含了未见过语言和语料库的额外数据。在规模上，训练数据是动态模拟的，包含了1000对验证数据以及900个盲测集中的语音片段。该任务的主要目标是语音增强。

This dataset contains noisy-clean audio pairs for training and validating speech enhancement models, covering a variety of distortion types such as packet loss, noise, reverberation, clipping, and more. The official blind test set consists of 50% real recordings and 50% synthetic data, and also includes additional data from unseen languages and corpora. In terms of scale, the training data is dynamically simulated, with 1000 validation pairs and 900 speech segments in the blind test set. The primary objective of this task is speech enhancement.