Collaborative Research: RI: Small: From ultrasound and MRI to articulatory and acoustic models of child speech development

成人语言的科学模型在几十年前就成熟了，并且在语音技术中被证明是非常有用的。相比之下，儿童语言模型仍然缺乏动机，并且基于成人而不是儿童的数据，部分原因是从儿童那里收集语音清晰度数据存在困难。随着儿童技术应用领域的出现和越来越重要，缺乏适当的儿童语言模型的问题日益严重。该项目将推进与儿童语言有关的最新科学技术，以开发各种应用，如用于识字发展和学习的基于语言的互动系统。从科学上讲，该项目将1)揭示儿童语言中发音和声学之间的关系，2)描述以前未被观察到的儿童语言发展过程，包括声道的解剖发育，以及3)开发儿童语言的声学模型和语音技术，这些模型和语音技术可以随着年龄的增长而扩展。该项目的不同方面是协同的：来自发音和声学实验的发现将为儿童语音技术的关键算法的发展提供信息。该项目的主要优势是：1)将创建和利用的数据语料库的独特性，以及2)跨学科的方法来研究语音产生和开发用于语音技术的儿童语音模型。生产数据将包括舌头的3D/4D超声记录、头部和颈部的MRI解剖扫描、上颚印痕和麦克风记录。语料库将首次允许发音和声学变化之间的关系，以及语音发音的纵向发展，直接探索和模拟儿童语音。该项目将促进对发音-声学关系及其在儿童语音中的发展的认识，以及它们与儿童语音的机器识别的关系。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Scientific models of adult speech matured several decades ago and have proven extremely useful in speech technologies. In contrast, models of child speech remain poorly motivated and grounded in adult rather than child data, in part because of the difficulties involved in collecting speech articulation data from children. The lack of appropriate models for child speech is increasingly problematic as technological application areas for children emerge and become more important. The project will advance the scientific and technological state-of-the-art related to child speech to develop a variety of applications such as speech-based interactive systems for literacy development and learning.Scientifically, the project will 1) reveal relationships between articulation and acoustics in child speech, 2) characterize previously unobserved processes of speech development in children, including anatomical development of the vocal tract, and 3) develop acoustic models and speech technologies for child speech, which can be scalable with age. The different aspects of the project are synergistic: findings from articulation and acoustic experiments will inform the development of algorithms essential to speech technologies for children. Key strengths of the project are the 1) uniqueness of the data corpus that will be created and leveraged, and 2) interdisciplinary approach to studying speech production and developing models of child speech for use in speech technologies. Production data will include 3D/4D ultrasound recordings of the tongue, MRI anatomical scans of the head and neck, palate impressions, and microphone recordings. The corpus will, for the first time, allow the relationship between articulatory and acoustic variability, and the longitudinal development of speech articulation, to be directly explored and modeled for child speech. The project will advance knowledge of articulatory-acoustic relations and their development in children's speech, and their relationship to machine recognition of children's speech.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.