Prolonging Functional Speech in Persons with Amyotrophic Lateral Sclerosis: A Real-Time Virtual Vocal Tract

肌萎缩侧索硬化症患者最终不可避免地会因进行性语言障碍而遭受严重的言语障碍 控制舌头、嘴唇和下巴运动的脑细胞退化。尽管 这种语言障碍对生活质量和生存的破坏性后果，几乎没有选择 可用于帮助受损的口头交流，以及许多现有的语音生成技术 运行缓慢，成本高昂。该项目旨在改善患有癌症的人的生活质量。 通过测试低成本语音生成设备(A)的有效性来测试ALS导致的语音受损 虚拟声道)，可以显著延长这些患者的口头交流能力。如果 如果成功，这些技术可以扩展到患者的广泛语音范围。 运动障碍。 虚拟声道使用机器学习算法来预测一个人想要做什么 比方说，仅仅基于嘴唇的动作，实时地。该设备的用户能够触发播放 通过简单地试图清楚地表达他们想要说的话，来识别一些预先确定的短语。我们的 以前的工作已经证明了使用成本过高的实验室系统的这种方法的可行性 例如电磁关节成像术。三维深度测绘相机技术的最新进展 允许使用低成本、可移植的技术首次测试这些技术 并且已经被集成到笔记本电脑和手机等消费设备中。 为此，正在开发的系统将在60名ALS患者中进行测试，这意味着 言语障碍的范围从正常到严重(正常15人，轻度15人，15人 中度，15例重度)。在测试过程中，参与者将被提示以随机的方式发音 在他们感到舒服的情况下，尽快下单。整个会议将被记录下来，并记录以下内容 将离线测量变量：识别准确率、识别延迟、任务时间、完成百分比、 和通信速率(每分钟字数)。用户将对该软件的可用性和接受性进行评分 在设备测试后立即使用虚拟声道，使用系统可用性量表。结果： 这项测试将用于解决以下具体目标：(1)确定准确性和 基于虚拟声道的构音障碍实时短语合成的延迟，(2) 确定使用虚拟声道产生的实时短语的可用性和可接受性， 以及(3)识别降低识别准确率的发音和语音因素。

People with ALS eventually and inevitably experience serious speech impairment due to progressive deterioration of brain cells that control movements of the tongue, lips and jaw. Despite the devastating consequences of this speech impairment on quality of life and survival, few options are available to assist impaired oral communication, and many existing speech-generating technologies are slow to operate and cost prohibitive. This project seeks to improve quality of life for persons with impaired speech due to ALS by testing the effectiveness of a low-cost, speech-generating device (a virtual vocal tract) that could significantly prolong the ability of these patients to communicate orally. If successful, these techniques could be extended for use by patients' with a broad range of speech motor impairments. The virtual vocal track uses machine learning algorithms to predict what a person is attempting to say, in real-time, based solely on lip movements. Users of the device are able to trigger the playback of a number of predetermined phrases by simply attempting to articulate what they want to say. Our previous work has shown the feasibility of this approach using cost-prohibitive laboratory systems such as electromagnetic articulography. Recent advances in 3D depth mapping camera technology allow these techniques to be tested for the first time using technologies, which are low-cost, portable and already being integrated into consumer devices such as laptops and cellphones. To this end, the system under development will be tested in 60 patients with ALS, representing a range of speech impairment from normal to severe speech intelligibility (15 normal, 15 mild, 15 moderate, 15 severe). During testing, participants will be cued to articulate the phrases in a random order as fast as is comfortable for them. The entire session will be recorded and the following variables will be measured offline: recognition accuracy, recognition latency, task time, % completion, and communication rate (words per minute). Users will rate the usability and acceptability of the virtual vocal tract immediately following device testing, using the System Usability Scale. Results of this testing will be used to address the following specific aims: (1) Determine the accuracy and latency of real-time phrase synthesis based on dysarthric speech using the virtual vocal tract, (2) Determine the usability and acceptability of real-time phrases produced using the virtual vocal tract, and (3) Identify the articulatory and speech factors that degrade recognition accuracy.