ACES: A Product to Suppress or Enhance Critical Components in Acoustic Signals

描述（由申请人提供）： 常见的声学环境通常是来自多个声源的声音的复杂混合。其中一些信息源包含了听众需要理解的关键信息;另一些则是干扰听众理解的干扰因素。随着美国人口的老龄化，越来越多的人难以应对如此复杂的声场。当前的解决方案限于众所周知地放大所有声源的助听器，或者选择性地放大单个源但被动地或主动地将收听者与他或她的声学环境的其余部分隔离的耳机。 我们建议开发一种名为ACES（Acoustic Component Enhancement System，声学组件增强系统）的产品，通过向听众呈现从其实际声学环境重建的虚拟声场来帮助他们，以这种方式增强（如果听众想要关注它们）或抑制（如果它们分散注意力）某些源2（称为可追踪源2）。可追踪声源是存在某种声学前信息或迹线的声源，其可用于识别和隔离声源的声音。为了隔离声音，ACES将使用一种新的基于知识的组件，称为源假设生成器（SHG）。星星已经确定了常见的可追踪源的重要类别，可以构建这种SHG。例如，扬声器产生的任何声音都是可追踪的。在这个重要的例子中，扬声器是声源，驱动扬声器的电信号是它的轨迹。 如果一个可追踪的声源是有用的，ACES会在ACES为听众构建的虚拟声场中创建一个增强版本。要做到这一点，ACES必须抑制原始的声学表示（可能会失真，难以理解），并将其替换为3更易于理解的4版本。ACES可以通过更大声地播放、时移、重复以及更慢或更快地播放来增强重建的声音。 星星在实现最先进的盲源分离算法方面拥有丰富的经验，可以从多个麦克风信号的混合响应中分离出独立的声源。在第一阶段，我们计划以新的方式扩展这些算法，以利用ACES声源假设所代表的独立的基于知识的信息。实际上，我们将3从物理麦克风对声场的响应中删除4可追踪源的贡献。其结果应该是在一个需要更少麦克风并更好地满足用户6需求的系统中，改进可追踪和不可追踪3隐藏4源的分离。我们还将进行感知实验，以测试在嘈杂环境中呈现的口语单词的可理解性，当可追踪的语音源已被如上所述地增强时，或者当其掩蔽噪声已被使用ACES技术抑制时。 我们预计，将ACES与辅助听力设备和助听器相结合，将需要三到四年的进一步发展，并需要100万至200万美元的支持资金。 公共卫生相关性： 拟议项目支持ACES（声学组件增强系统）的开发，这是一种提高听力受损听众生活质量的产品，包括许多老年婴儿潮一代：这些听众难以处理复杂的声学环境，其中一些声源包含他们需要理解的关键信息，其他来源是干扰理解的干扰。ACES使用新的过滤技术来抑制分散注意力的来源，并增强信息承载的来源（例如，公告）。ACES采用这些技术来生成一个由用户控制的“虚拟现实”，在这个虚拟现实中，干扰被抑制，重要的声音被增强。

DESCRIPTION (provided by applicant): Common acoustic environments are often complex mixtures of sounds from multiple acoustic sources. Some of these sources contain critical information listeners need to comprehend; others are distractions that interfere with listeners6 comprehension. As the US population ages, a significant and growing segment have difficulty coping with such complex sound fields. Current solutions are limited to hearing aids which notoriously amplify all acoustic sources, or headsets which selectively amplify a single source but passively or actively isolate the listener from the rest of his or her acoustic environment. We propose to develop a product called ACES (Acoustic Component Enhancement System) to help listeners by presenting them with a virtual sound field reconstructed from their actual acoustic environment in such a way that certain sources2called traceable sources2are enhanced (if they listener wants to attend to them) or suppressed (if they are distracting). Traceable sources are acoustic sources for which some kind of pre-acoustic information, or trace, exists that can be used to identify and isolate the sound of the source. To isolate the sound, ACES will use a novel knowledge-based component called a Source Hypothesis Generator (SHG). STAR has identified important classes of common traceable sources for which such SHGs can be constructed. For instance, any sound produced by a loudspeaker is traceable. In this important case, the speaker is the acoustic source, and the electrical signal that drives the speaker is its trace. If a traceable source is informative, ACES creates an enhanced version of it in the virtual sound field that ACES constructs for the listener. To do so, ACES must suppress the original acoustic representation (which may be distorted and difficult to comprehend) and replace it with a more 3listener-friendly4 version. ACES can enhance the reconstructed sound by playing it louder, time-shift it, repeat it, and play it slower or faster. STAR has extensive experience implementing state-of-the-art Blind Source Separation algorithms to separate independent acoustic sources from the mixed responses of multiple microphone signals. In Phase I, we plan to extend those algorithms in novel ways to leverage the independent knowledge-based information represented by the ACES acoustic source hypotheses. In effect, we will 3scrub4 the contributions of the traceable sources out of the physical microphone responses to the sound field. The result should be improved separation of both traceable and un-traceable 3hidden4 sources, in a system that requires fewer microphones, and meets users6 needs better. We will also carry out a perceptual experiment to test the intelligibility of spoken words presented in a noisy environment, when the traceable speech source has been enhanced as described above, or when its masking noises have been suppressed using the ACES technology. We anticipate that integrating ACES with assistive listening devices and hearing aids will require three or four years of further development beyond this Phase I project, and require $1-2 million in supporting funds. PUBLIC HEALTH RELEVANCE: The proposed project supports the development of ACES (Acoustic Component Enhancement System), a product to enhance the quality of life of hearing-impaired listeners, including many aging baby-boomers: such listeners have difficulty processing complex acoustic environments in which some acoustic sources contain critical information they need to comprehend, and other sources are distractions that interfere with comprehension. ACES uses novel filtering techniques to suppress distracting sources and enhance information-bearing ones (for example, announcements). ACES employs these techniques to generate a listener-controlled "virtual reality" in which distractions are suppressed and important sounds are enhanced.