Speech Perception under Cognitive Load

认知负荷下的言语感知

• 批准号: ES/R004722/1
• 负责人: Sven Mattys
• 金额: $ 36.25万
• 依托单位: University of York
• 依托单位国家: 英国
• 项目类别: Research Grant
• 财政年份: 2018
• 资助国家: 英国
• 起止时间: 2018 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=ES%2FR004722%2F1
• 关键词: Speech; Perception; under; Cognitive; Load

Most theories of human speech perception are derived from tasks performed in a quiet environment and under conditions of undivided attention. However, in the past few years, there has been a surge of interest in modelling speech recognition in more realistic conditions (e.g., noisy background, accented speech). However, among these realistic conditions, those resulting from a cognitive load have received little attention. Here, we define cognitive load (CL) as any listening challenges arising not from a distortion of the speech signal but from the recruitment of processing resources due to concurrent attentional or mnemonic demands. For example, what are the consequences of monitoring cockpit instruments on a pilot's ability to follow spoken instructions from ground control? The disruptive effect of CL on speech perception is noticed as early as in the initial stages of acoustic encoding. Under some circumstances, CL can even lead to a form of transient hearing impairment called inattentional deafness. Despite the obvious implications that these results have for theory and clinical practice, little is known about the low-level mechanisms by which CL interferes with speech perception. The aim of this proposal is to address this issue in three interconnected research streams drawing upon psychometric and identification paradigms.The first stream asks whether CL affects all acoustic dimensions of speech equally. This question is important because not all acoustic dimensions are equally crucial for communication. For example, successful word recognition is more resilient to pitch distortions than duration distortions. The idea that CL affects some dimensions more than others is motivated by the claim that CL (e.g., a concurrent visual task) causes listeners to rapidly shift attention back and forth between the speech signal and the CL task, leading to an underestimation of the duration of the speech signal. If this hypothesis is correct, CL should lead primarily to a distortion of auditory temporal judgements and leave other core dimensions (loudness, pitch, and spectral structure) unaffected. This will be contrasted with the claim that CL leads to a general reduction in auditory precision across all acoustic dimensions.The second stream investigates whether the format of the CL stimuli affects the severity of the CL interference. For example, is speech perception more affected by a concurrent task that requires rehearsing words silently (phonological format) or by a task that requires processing visual stimuli (visual format)? These experiments will address the debate between modal and amodal views of the processing resources used during speech perception. The third stream aims to distinguish two potential mechanisms behind CL interference: Encoding and maintenance. Encoding is the process of converting a sensory input into mental representations. Maintenance is the process of preserving these representations in memory. Encoding of the CL stimuli will be manipulated such that it takes place either during or before the speech stimuli, hence pitting encoding against maintenance as the mechanism underlying interference. An encoding hypothesis predicts that only simultaneous encoding of speech and CL stimuli should lead to CL effects. In order to explore the generalisability of the above phenomena beyond the speech domain, the effect of CL will be tested on both speech and non-speech sounds. This comparison will situate our findings within the long-standing debate on the existence of a specialised speech mode for sound perception.Finally, because the notion of "cognitive listening" is becoming central not only in speech research but also in hearing practice, we will engage with clinical audiologists and discuss ways of including a cognitive component into standard pure-tone audiometric (PTA) and advise on potential phase-II clinical trials.

大多数人类语音感知的理论都是源自在安静环境和全神贯注的条件下执行的任务。然而，在过去几年中，人们对在更现实的条件下（例如，嘈杂的背景、带口音的语音）下的语音识别建模产生了浓厚的兴趣。然而，在这些现实情况中，那些由认知负荷引起的情况却很少受到关注。在这里，我们将认知负荷（CL）定义为不是由语音信号失真引起的任何听力挑战，而是由于同时存在的注意力或记忆需求而需要处理资源而引起的。例如，监控驾驶舱仪器对飞行员遵循地面控制语音指令的能力有何影响？早在声学编码的初始阶段就注意到了 CL 对语音感知的破坏性影响。在某些情况下，CL 甚至可能导致一种称为注意力不集中性耳聋的短暂性听力障碍。尽管这些结果对理论和临床实践具有明显的影响，但人们对 CL 干扰言语感知的低级机制知之甚少。该提案的目的是利用心理测量和识别范式在三个相互关联的研究流中解决这个问题。第一个流询问 CL 是否平等地影响语音的所有声学维度。这个问题很重要，因为并非所有声学维度对于通信都同样重要。例如，成功的单词识别对音调失真的抵抗力比对时长失真的抵抗能力更强。 CL 对某些维度的影响比其他维度更大的想法是由于 CL（例如，并发视觉任务）导致听众在语音信号和 CL 任务之间快速来回转移注意力，从而导致对语音信号持续时间的低估。如果这个假设是正确的，CL 应该主要导致听觉时间判断的失真，并且其他核心维度（响度、音调和频谱结构）不受影响。这与 CL 导致所有声学维度的听觉精度普遍降低的说法形成鲜明对比。第二个流研究 CL 刺激的格式是否影响 CL 干扰的严重程度。例如，言语感知是否更容易受到需要默默排练单词（语音格式）的并发任务的影响，还是受到需要处理视觉刺激（视觉格式）的任务的影响？这些实验将解决语音感知过程中使用的处理资源的模态和非模态视图之间的争论。第三个流旨在区分 CL 干扰背后的两种潜在机制：编码和维护。编码是将感觉输入转换为心理表征的过程。维护是将这些表示形式保存在内存中的过程。 CL 刺激的编码将被操纵，使其在语音刺激期间或之前发生，从而使编码与维护成为干扰的潜在机制。编码假设预测，只有语音和 CL 刺激的同时编码才会导致 CL 效应。为了探索上述现象在语音领域之外的普遍性，将在语音和非语音上测试 CL 的效果。这种比较将把我们的研究结果置于关于声音感知的专门语音模式是否存在的长期争论中。最后，由于“认知听力”的概念不仅在语音研究中而且在听力实践中变得中心，我们将与临床听力学家合作，讨论将认知成分纳入标准纯音听力测定（PTA）的方法，并就潜在的 II 期临床试验提供建议。