ISOLET数据集 150名受试者两次说出字母表中每个字母的名称

Data Set Information: This data set was generated as follows. 150 subjects spoke the name of each letter of the alphabet twice. Hence, we have 52 training examples from each speaker. The speakers are grouped into sets of 30 speakers each, and are referred to as isolet1, isolet2, isolet3, isolet4, and isolet5. The data appears in isolet1+2+3+4.data in sequential order, first the speakers from isolet1, then isolet2, and so on. The test set, isolet5, is a separate file. You will note that 3 examples are missing. I believe they were dropped due to difficulties in recording. I believe this is a good domain for a noisy, perceptual task. It is also a very good domain for testing the scaling abilities of algorithms. For example, C4.5 on this domain is slower than backpropagation! I have formatted the data for C4.5 and provided a C4.5-style names file as well. Attribute Information: The features are described in the paper by Cole and Fanty cited above. The features include spectral coefficients; contour features, sonorant features, pre-sonorant features, and post-sonorant features. Exact order of appearance of the features is not known. Relevant Papers: Fanty, M., Cole, R. (1991). Spoken letter recognition. In Lippman, R. P., Moody, J., and Touretzky, D. S. (Eds). Advances in Neural Information Processing Systems 3. San Mateo, CA: Morgan Kaufmann. [Web link] Dietterich, T. G., Bakiri, G. (1991) Error-correcting output codes: A general method for improving multiclass inductive learning programs. Proceedings of the Ninth National Conference on Artificial Intelligence (AAAI-91), Anaheim, CA: AAAI Press. [Web link] Dietterich, T. G., Bakiri, G. (1994) Solving Multiclass Learning Problems via Error-Correcting Output Codes. Available as URL: [Web link] [Web link] Papers That Cite This Data Set1: Jaakko Peltonen and Samuel Kaski. Discriminative Components of Data. IEEE. 2004. [View Context]. Vassilis Athitsos and Stan Sclaroff. Boosting Nearest Neighbor Classifiers for Multiclass Recognition. Boston University Computer Science Tech. Report No, 2004-006. 2004. [View Context]. David Littau and Daniel Boley. Using Low-Memory Representations to Cluster Very Large Data Sets. SDM. 2003. [View Context]. Inderjit S. Dhillon and Dharmendra S. Modha and W. Scott Spangler. Class visualization of high-dimensional data with applications. Department of Computer Sciences, University of Texas. 2002. [View Context]. Erin L. Allwein and Robert E. Schapire and Yoram Singer. Reducing Multiclass to Binary: A Unifying Approach for Margin Classifiers. ICML. 2000. [View Context]. Creators: Ron Cole and Mark Fanty Department of Computer Science and Engineering, Oregon Graduate Institute, Beaverton, OR 97006. cole '@' cse.ogi.edu, fanty '@' cse.ogi.edu Donor: Tom Dietterich Department of Computer Science Oregon State University, Corvallis, OR 97331 tgd '@' cs.orst.edu

数据集信息：本数据集生成方式如下。150名受试者每人将字母表中每个字母的名称朗读两次，因此每位说话者可提供52条训练样本。所有说话者按每30人一组划分为5组，分别记为isolet1、isolet2、isolet3、isolet4与isolet5。数据集按顺序存储于isolet1+2+3+4.data文件中，依次包含isolet1组、isolet2组直至isolet4组的受试者数据；测试集isolet5单独存储为独立文件。需注意的是，数据集中共缺失3条样本，推测是因录制过程中出现问题而被移除。本数据集非常适合用于含噪声的感知任务研究，同时也是测试算法缩放性能的优质测试域。例如，在该数据集上C4.5的运行速度慢于反向传播（backpropagation）算法。本数据集已针对C4.5算法完成格式适配，并附带了C4.5格式的属性名称文件。

属性信息：数据集的特征说明可参考前文引用的Cole与Fanty的研究论文。特征包含频谱系数、轮廓特征、浊音特征、前浊音特征与后浊音特征。目前暂未明确特征的具体出现顺序。

相关论文：
Fanty, M., Cole, R. (1991). 语音字母识别。收录于Lippman, R. P., Moody, J., 与Touretzky, D. S. 主编的《神经信息处理系统进展3》（Advances in Neural Information Processing Systems 3），加利福尼亚州圣马特奥：Morgan Kaufmann出版社。[网页链接]
Dietterich, T. G., Bakiri, G. (1991) 纠错输出码：一种改进多分类归纳学习程序的通用方法。发表于第九届全国人工智能大会（AAAI-91）会议论文集，加利福尼亚州阿纳海姆：AAAI出版社。[网页链接]
Dietterich, T. G., Bakiri, G. (1994) 基于纠错输出码求解多分类学习问题。可通过以下网址获取：[网页链接] [网页链接]

引用本数据集的论文：
Jaakko Peltonen与Samuel Kaski. 数据的判别分量. IEEE. 2004. [查看上下文]
Vassilis Athitsos与Stan Sclaroff. 提升近邻分类器用于多分类识别. 波士顿大学计算机科学技术报告第2004-006号. 2004. [查看上下文]
David Littau与Daniel Boley. 采用低内存表示方法对超大规模数据集进行聚类. SDM. 2003. [查看上下文]
Inderjit S. Dhillon、Dharmendra S. Modha与W. Scott Spangler. 高维数据的类别可视化及应用. 德克萨斯大学计算机科学系. 2002. [查看上下文]
Erin L. Allwein、Robert E. Schapire与Yoram Singer. 将多分类问题归约为二分类：一种面向边际分类器的统一方法. ICML. 2000. [查看上下文]

数据集创建者：Ron Cole与Mark Fanty
俄勒冈研究生院计算机科学与工程系，比弗顿，俄勒冈州97006。
联系邮箱：cole '@' cse.ogi.edu，fanty '@' cse.ogi.edu

数据集提供者：Tom Dietterich
俄勒冈州立大学计算机科学系，科瓦利斯，俄勒冈州97331
联系邮箱：tgd '@' cs.orst.edu

搜集汇总

背景与挑战

背景概述

ISOLET数据集是一个语音识别数据集，包含150名受试者两次说出字母表中每个字母的名称，生成52个训练样本。数据集分为五个组，测试集单独存放，适用于语音识别和多类分类任务的研究。