多说话人、多语句语音分离数据集
收藏资源简介:
该数据集由芬兰坦佩雷大学信号处理研究中心和芬兰诺基亚技术合作创建,用于评估未知说话人数的多语句语音分离模型。数据集包含20,000个混合语音信号,每个信号由两个或三个说话人贡献多个语句。数据集在无回声、噪声、回声以及噪声和回声混合的条件下生成,以模拟真实世界的语音分离场景。数据集旨在解决未知说话人数和多个语句的语音分离问题,为相关研究提供实验数据。
Co-created by the Signal Processing Research Center of Tampere University (Finland) and Nokia Technologies Finland, this dataset is designed for evaluating speech separation models that handle multiple utterances with unknown speaker counts. It contains 20,000 mixed speech signals, each of which includes multiple utterances contributed by two or three speakers. The dataset is generated under four acoustic conditions: anechoic, noisy, reverberant, and combined noisy-reverberant scenarios, to simulate real-world speech separation environments. This dataset aims to address the speech separation problem involving unknown speaker counts and multiple utterances, providing experimental data for relevant research.

- 1Attractor-Based Speech Separation of Multiple Utterances by Unknown Number of Speakers芬兰坦佩雷大学信号处理研究中心, 芬兰诺基亚技术 · 2025年



