Singer Traits Dataset (STraDa)

STraDa是一个专注于歌唱声音分析的大型数据集，由法国Deezer公司创建。该数据集包含两个子集：automatic-strada自动生成，包含25,194条音频片段，覆盖25种音乐类型和35种语言；annotated-strada手工标注，包含200条平衡性别、年龄和语言的音频片段。数据集的创建过程涉及从多个公开音乐百科中匹配和处理数据，确保每个音频片段仅由单一主唱演唱。STraDa主要用于歌手性别分类、歌手识别和年龄检测等任务，旨在通过丰富的元数据和可下载的音频文件，提升模型性能并进行偏差分析。

STraDa is a large-scale dataset dedicated to singing voice analysis, developed by Deezer, a French company. It comprises two subsets: automatic-strada and annotated-strada. The automatic-strada subset, which is automatically generated, contains 25,194 audio clips spanning 25 music genres and 35 languages. The annotated-strada subset, which is hand-annotated, includes 200 audio clips balanced across gender, age and language groups. The dataset was constructed by matching and processing data from multiple public music encyclopedias, ensuring that each audio clip features exactly one single lead vocalist. STraDa is primarily applied to tasks such as singer gender classification, singer identification and age detection, with the objective of improving model performance and conducting bias analysis through its rich metadata and downloadable audio files.