amine-maazizi/tta-detection-2k
收藏资源简介:
--- pretty_name: TTA Deepfake Outputs + Ground Truth language: - en license: cc-by-4.0 multilinguality: monolingual task_categories: - audio-classification - text-to-audio task_ids: - multi-class-classification size_categories: - 1K<n<10K tags: - audio - deepfake-detection - text-to-audio - diffusion - audioldm - synthetic-audio - forensics configs: - config_name: audioldm1_manifest data_files: - split: train path: outputs/manifests/audioldm1_generation_manifest.csv - config_name: audioldm2_manifest data_files: - split: train path: outputs/manifests/audioldm2_generation_manifest.csv - config_name: real_subset_metadata data_files: - split: train path: data/tta_subset_full/real_subset_metadata.csv - config_name: clotho_captions_dev data_files: - split: train path: data/clotho_v2.1/clotho_captions_development.csv - config_name: clotho_captions_eval data_files: - split: test path: data/clotho_v2.1/clotho_captions_evaluation.csv dataset_info: - config_name: audioldm1_manifest features: - name: sample_id dtype: string - name: caption dtype: string - name: split dtype: string - name: gt_audio_path dtype: string - name: requested_duration dtype: float64 - name: model_name dtype: string - name: seed dtype: int64 - name: ddim_steps dtype: int64 - name: duration dtype: float64 - name: device dtype: string - name: prompt_text dtype: string - name: output_path dtype: string - name: status dtype: string - name: runtime_sec dtype: float64 - name: log_path dtype: string - config_name: audioldm2_manifest features: - name: sample_id dtype: string - name: caption dtype: string - name: split dtype: string - name: gt_audio_path dtype: string - name: requested_duration dtype: float64 - name: model_name dtype: string - name: seed dtype: int64 - name: ddim_steps dtype: int64 - name: duration dtype: float64 - name: device dtype: string - name: prompt_text dtype: string - name: output_path dtype: string - name: status dtype: string - name: runtime_sec dtype: float64 - name: log_path dtype: string ---
友好名称:TTA深度伪造音频输出与基准真值(Ground Truth)数据集 语言:英语 许可证:CC BY 4.0 多语言属性:单语言 任务类别: - 音频分类 - 文本转音频 任务子类别:多分类 数据规模:1000 < 样本数量 < 10000 标签: - 音频 - 深度伪造检测 - 文本转音频 - 扩散模型(diffusion) - AudioLDM(audioldm) - 合成音频 - 取证科学 配置方案: - 配置名称:AudioLDM v1 配置索引 数据文件: - 拆分集:训练集 路径:outputs/manifests/audioldm1_generation_manifest.csv - 配置名称:AudioLDM v2 配置索引 数据文件: - 拆分集:训练集 路径:outputs/manifests/audioldm2_generation_manifest.csv - 配置名称:真实子集元数据 数据文件: - 拆分集:训练集 路径:data/tta_subset_full/real_subset_metadata.csv - 配置名称:Clotho 开发集字幕 数据文件: - 拆分集:训练集 路径:data/clotho_v2.1/clotho_captions_development.csv - 配置名称:Clotho 评估集字幕 数据文件: - 拆分集:测试集 路径:data/clotho_v2.1/clotho_captions_evaluation.csv 数据集信息: - 配置名称:AudioLDM v1 配置索引 特征字段: 1. 样本ID:字符串类型 2. 字幕:字符串类型 3. 数据集拆分:字符串类型 4. 基准真值(Ground Truth)音频路径:字符串类型 5. 请求生成时长:64位浮点型 6. 模型名称:字符串类型 7. 随机种子:64位整型 8. DDIM采样步数:64位整型 9. 实际生成时长:64位浮点型 10. 运行设备:字符串类型 11. 提示文本:字符串类型 12. 输出文件路径:字符串类型 13. 生成状态:字符串类型 14. 运行耗时(秒):64位浮点型 15. 日志文件路径:字符串类型 - 配置名称:AudioLDM v2 配置索引 特征字段与上述AudioLDM v1配置索引完全一致



