LibriConvo
收藏资源简介:
LibriConvo 是一个基于语音感知会话模拟(SASC)的模拟多说话人对话数据集,旨在支持说话人分割和自动语音识别(ASR)系统的训练和评估。与先前主要依赖语义不连贯的语句和不合理的时间间隔的资源不同,LibriConvo 确保语义连贯性和现实的对话时间。该数据集包含 240.1 小时的音频,涵盖 1,496 个对话,涉及 830 个独特的说话人,并以说话人非重叠的方式分割,以便进行稳健的评估。
LibriConvo is a simulated multi-speaker conversation dataset based on Speech-Aware Session Simulation (SASC), designed to support the training and evaluation of speaker diarization and automatic speech recognition (ASR) systems. Unlike previous resources that primarily rely on semantically incoherent utterances and unrealistic temporal intervals, LibriConvo ensures semantic coherence and realistic conversational timing. This dataset contains 240.1 hours of audio, covering 1,496 conversations involving 830 unique speakers, and is segmented in a speaker-non-overlapping manner to enable robust evaluation.




