ABSTRACT

There are over 466 million people in the world with disabling hearing loss. People with severe-to-profound hearing impairment need to lipread or use sign language, even with hearing aids. Assistive Technologies play a vital role in helping these people interact efficiently with their environment. Deaf drivers are not currently able to take full advantage of voice-based navigation applications. In this paper, we describe research that is aimed at developing an assistive device that (1) recognizes voice-stream navigation instructions from GPS-based navigation applications, and (2) maps each voiced navigation instruction to a vibrotactile stimulus that can be perceived and understood by deaf drivers. A 13-element feature vector is extracted from each voice stream, and classified into one of six categories, where each category represents a unique navigation instruction. The classification of the feature vectors is done using a K-Nearest-Neighbor classifier (with an accuracy of 99.05%) which was found to outperform five other classifiers. Each category is then mapped to a unique vibration pattern, which drives vibration motors in real time. A usability study was conducted with ten participants. Three different alternatives were tested, to find the best body locations for mounting the vibration motors. The solution ultimately chosen was two sets of five vibrator motors, where each set was mounted on a bracelet. Ten drivers were asked to rate the proposed device (based on eight different factors) after they used the assistive device on 8 driving routes. The overall mean rating across all eight factors was 4.67 (out of 5) This indicates that the proposed assistive device was seen as useful and effective.

摘要

全世界有超过 4.66 亿人患有致残性听力损失。重度至极重度听力障碍的人需要阅读唇语或使用手语，即使有助听器也是如此。辅助技术在帮助这些人与环境有效互动方面发挥着至关重要的作用。聋人司机目前无法充分利用基于语音的导航应用程序。在本文中，我们描述了旨在开发一种辅助设备的研究，该辅助设备 (1) 识别来自基于 GPS 的导航应用程序的语音流导航指令，以及 (2) 将每个语音导航指令映射到可以感知和聋司机都能理解。从每个语音流中提取一个 13 元素的特征向量，并分为六类之一，其中每个类别代表一个独特的导航指令。特征向量的分类是使用 K-Nearest-Neighbor 分类器（准确率为 99.05%）完成的，该分类器的性能优于其他五个分类器。然后将每个类别映射到一个独特的振动模式，实时驱动振动电机。对 10 名参与者进行了可用性研究。测试了三种不同的替代方案，以找到安装振动电机的最佳车身位置。最终选择的解决方案是两组，每组五个振动电机，每组都安装在手镯上。在 8 条驾驶路线上使用辅助设备后，要求 10 名司机对提议的设备（基于 8 个不同的因素）进行评分。所有八个因素的总体平均评分为 4。