ivkond/synthetic-speech-diarization-ru

这是一个合成多说话人音频数据集，用于说话人对话任务的训练和评估。每个音轨包含2-4个说话人的多说话人对话，具有精确的时间戳和转录的说话人片段，多种对话模式（对话、独白、小组讨论、访谈），以及现实特征（重叠、同时说话、背景噪声）。数据集分为容易、中等、困难三个难度级别。

A synthetic multi-speaker audio dataset for speaker diarization tasks, generated from the FBK-MT/Speech-MASSIVE-test dataset. Each track contains multi-speaker conversations with 2-4 speakers, speaker segments with precise timestamps and transcriptions, various conversation patterns (dialogues, monologues, group discussions, interviews), realistic features (overlaps, simultaneous speech, background noise), and difficulty levels (easy, medium, hard).