Brennan M, Kirkland J
Percept Mot Skills. 1979 Jun;48(3 Pt 1):683-6. doi: 10.2466/pms.1979.48.3.683.
A technique for clustering infants' cry-signals on the basis of perceptually discriminable acoustic characteristics is reported. A cry-sequence containing all pairwise combinations of 24 signals was constructed. These signals have been used previously in cry-recognition studies (e.g., Wasz-Hockert, et al., 1968) and consist of six each of pain, hunger, pleasure, and birth. 20 musically competent subjects were tested individually and instructed to compare each cry with the one immediately preceding it in the sequence. Forced-choice "similarity" scores were summed for every pair of cries. These "similarity scores" were data for both a manual and a computerized clustering method. There was absolute agreement between these two methods, and the clusters of cry-signals correspond very closely to the four original cry groups. This seems a valid technique for examining the acoustic similarity of infants' cry-signals.
本文报道了一种基于可感知的声学特征对婴儿哭声信号进行聚类的技术。构建了一个包含24种信号所有两两组合的哭声序列。这些信号先前已用于哭声识别研究（例如，瓦兹 - 霍克特等人，1968年），包括六种疼痛、饥饿、愉悦和出生时的哭声。20名具有音乐能力的受试者分别接受测试，并被要求将序列中的每一声哭声与紧接其前的哭声进行比较。对每对哭声的强制选择“相似度”得分进行求和。这些“相似度得分”是手动聚类方法和计算机化聚类方法的数据。这两种方法之间存在完全一致性，并且哭声信号的聚类与四个原始哭声组非常紧密地对应。这似乎是一种检验婴儿哭声信号声学相似性的有效技术。
