Listener sensitivity to talker differences in phonetic properties of speech

描述(由申请人提供)：拟议研究的目标是扩展我们对单词识别早期阶段的知识，在该阶段中，听者从语音信号中提取单独的片段。长期以来对言语知觉的描述强调语言表征的抽象性，但最近的研究结果表明，特定于说话者的声学-语音信息保留在记忆中，可以促进单词识别，这一点受到了挑战。这些发现提出了一种可能性，即详细的声学-语音信息被用于在特定说话者的基础上定制信号和分段表示之间的映射。为了支持这一替代说法，现在有证据表明，听众可以跟踪特定说话者的声学-语音特性。一个这样的属性是语音开始时间(VOT)，这是语音的一种时间属性，它标志着停止辅音中的发音对比。听众可以在一个单词的背景下学习说话者的特征投票权--首个无声的停顿，而且，还可以将这些信息转移到以相同停顿开头的新词上。一个基本问题仍然没有得到回答，这涉及到跟踪特定说话者的声学-语音信息的表征水平。这项拟议的研究的具体目的是通过确定听者是否跟踪说话者特定的VOT来解决这个问题，这些VOT是关于语音特征还是关于给定的语音片段。在培训阶段，听者将学习两个说话者如何产生/p/或/k/。语音合成技术将被用来操纵两个说话者的投票，使一个说话者的投票数较短，而另一个说话者的投票数较长。在测试阶段，将使用两种选择的强迫选择任务来检查对以训练中使用的相同清音停顿开头的单词的迁移，以及对以不同发音位置的清音停顿开头的单词的迁移。如果听话者根据语音特征跟踪说话者特定的Vot，那么在/p/语境中学到的关于无声辅音的信息应该转移到/k/，并且在/k/语境中学到的信息应该转移到/p/。然而，如果听话者跟踪特定于说话者的特定VOT，那么转移应该仅限于以训练期间使用的相同无声停顿开头的单词。这一研究将有助于从理论上理解说话者在言语知觉中的专一性，并为识别正常和混乱语音的设备的发展提供支持。这种设备目前的一个局限性是不能快速适应说话者在语音产生中的差异。研究人类如何处理这种类型的语音变化将提供关键信息，以便纳入到口语的机器识别中。

DESCRIPTION (provided by applicant): The goal of the proposed research is to extend our knowledge of the early stages of word recognition in which listeners extract individual segments from the speech signal. Long-standing accounts of speech perception, which emphasized the abstract nature of linguistic representations, have recently been challenged by findings that indicate that talker-specific, acoustic-phonetic information is retained in memory and can facilitate word recognition. These findings raise the possibility that detailed acoustic-phonetic information is used to customize the mapping between signal and segmental representation on a talker- specific basis. In support of this alternative account, there is now evidence that listeners can track acoustic- phonetic properties for a particular talker. One such property is voice-onset-time (VOT), a temporal property of speech that marks the voicing contrast in stop consonants. Listeners can learn a talker's characteristic VOTs in the context of one word-initial voiceless stop and, moreover, can transfer this information to a novel word that begins with the same stop. A fundamental question that remains unanswered concerns the level of representation at which talker-specific, acoustic-phonetic information is tracked. The specific aim of the proposed research is to address this question by determining whether listeners track talker-specific VOT with respect to a phonetic feature or with respect to a given phonetic segment. During a training phase, listeners will learn how two talkers produce /p/ or /k/. Speech synthesis techniques will be used to manipulate the VOTs of the two talkers so that one talker has shorter VOTs and the other talker has longer VOTs. During a test phase, a two-alternative forced-choice task will be used to examine transfer to words that begin with the same voiceless stop as used during training and to words that begin with a voiceless stop at a different place of articulation. If listeners track talker-specific VOT with respect to a phonetic feature, then information learned about voiceless stop consonants in the context of /p/ should transfer to /k/ and information learned in the context of /k/ should transfer to /p/. However, if listeners track talker-specific VOT with respect to a given phonetic segment, then transfer should be limited only to words that begin with the same voiceless stop as used during training. This research will contribute to the theoretical understanding of talker specificity in speech perception as well as support the advancement of devices that recognize normal and disordered speech. One current limitation of such devices is the failure to rapidly adapt to talker differences in speech production. Examining how humans process this type of phonetic variation will provide critical information to incorporate into machine recognition of spoken language.