Perceptimatic English Benchmark (PEB)
收藏arXiv2020-05-07 更新2024-08-06 收录
下载链接：
http://arxiv.org/abs/2005.03418v1
下载链接
链接失效反馈官方服务：
资源简介：
Perceptimatic English Benchmark (PEB) 是由巴黎大学的研究团队创建的一个开放实验基准，用于评估英语语音感知的定量模型。该数据集包含ABX刺激以及91名美国英语听众的反应，测试了大量英语和法语音位对比的辨别能力。数据集的刺激直接从朗读语料库中提取，适合评估训练在典型语音数据集上的统计声学模型。PEB数据集旨在解决人类语音感知的计算模型与实际语音之间的差异问题，特别是在语音识别领域。
Perceptimatic English Benchmark (PEB) is an open experimental benchmark developed by a research team at the University of Paris for evaluating quantitative models of English speech perception. This dataset contains ABX stimuli and listener responses from 91 American English listeners, and tests the discriminative ability of a large number of English and French phonemic contrasts. The stimuli in the dataset are directly extracted from read speech corpora, making it suitable for evaluating statistical acoustic models trained on typical speech datasets. The PEB dataset is designed to address the discrepancy between computational models of human speech perception and real-world speech, particularly in the field of speech recognition.
提供机构：
巴黎大学，LLF，CNRS，巴黎，法国CoML，ENS/CNRS/EHESS/INRIA/PSL研究大学，巴黎，法国CRI，Frontieres du Vivant et de l'Apprendre系，IIFR，巴黎大学
创建时间：
2020-05-07
