AudioMarkData
收藏资源简介:
AudioMarkData是由杜克大学创建的一个音频数据集,旨在评估音频水印技术的鲁棒性。该数据集包含20,000个音频样本,来源于Common Voice数据集,涵盖25种语言、不同生物性别和年龄段,确保了多样性和平衡性。创建过程中,数据集不仅提供了水印和非水印音频,还包括了多种扰动下的音频样本,以便于评估水印技术的有效性。该数据集主要应用于音频水印技术的鲁棒性测试,特别是在对抗水印移除和水印伪造的扰动方面,为解决合成音频的真实性问题提供了重要资源。
AudioMarkData is an audio dataset developed by Duke University, designed to evaluate the robustness of audio watermarking technologies. This dataset contains 20,000 audio samples sourced from the Common Voice dataset, covering 25 languages, diverse genders and age groups, ensuring diversity and balance. During its development, the dataset provides not only watermarked and unwatermarked audio samples but also audio samples under various perturbations, facilitating the evaluation of the effectiveness of watermarking technologies. This dataset is primarily used for robustness testing of audio watermarking technologies, especially against perturbations such as watermark removal and watermark forgery, serving as a critical resource for addressing the authenticity issue of synthetic audio.

- 1AudioMarkBench: Benchmarking Robustness of Audio Watermarking杜克大学 · 2024年



