SBIR Phase I: Touch-free Voice Trigger for Power Constrained Systems

这项小企业创新研究（SBIR）第一阶段项目的更广泛影响/商业潜力将是通过智能语音技术为经济、社会和科学/技术理解带来积极影响的变化。在商业上，智能、超低功耗、嵌入式语音触发器的推出将在1130亿美元的全球语音和语音识别市场中占据重要份额，有助于其扩张以及可穿戴设备和物联网（IoT）等相关市场的扩张。长期的潜在影响是广泛的，包括语音识别生态系统中的关键参与者，如移动电话行业、半导体公司、芯片设计师、技术提供商和系统集成商，他们将能够通过拟议的平台进行创新。在社交方面，通过更快、更自然地使用语音命令来访问设备功能并实时检索处理过的信息，用户的日常生活将变得更加轻松。此外，通过执行车载语音处理，该技术通过基于云的解决方案缓解了围绕隐私和数据保护的许多社会问题和挑战。这个小型企业创新研究第一阶段项目旨在开发一种超低功耗的永开、免触摸嵌入式语音触发器，用于部署在智能产品和物联网系统中。这种新颖的语音处理技术将模拟语音信号转换为生物启发的数据高效二进制脉冲序列，并使用自适应状态模型对其进行处理。这些模型可以部署在可重新编程的自动机网络中，具有高精度，仅使用由用户声音产生的数字脉冲的定时信息。与传统的语音特征相比，脉冲序列在人类听觉系统中使用，对噪声具有很强的鲁棒性，并且导致非常小的硬件实现，非常适合低功耗片上处理。此外，通过使用自动机仅处理1和0，所提出的方法不需要算法，从而实现实时、复杂的语音处理能力，否则将以高昂的成本实现或降级到云端。这种关键字定位技术是预先可配置的，为每个用户定制，可用于手机或类似的便携式设备，校准后不需要网络连接。它通过消除持续在线发送潜在敏感数据的需要，为访问便携式设备功能提供了更高的安全性和隐私性。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project will be to bring positive impactful changes to the economy, society, and scientific/technological understanding through intelligent voice-enabled technology. Commercially, the introduction of an intelligent, ultra-low power, embedded voice trigger will capture an important portion of the $113 billion global voice and speech recognition market, contributing to its expansion and that of related markets such as wearables and Internet of Things (IoT). The long-term potential impact is broad and encompasses the key players in the voice recognition ecosystem such as the mobile phone industry, semiconductor companies, chip designers, technology providers and system integrators, who will be enabled to innovate with the proposed platform. Societally, the day-to-day lives of users will be made easier by empowering them through the faster and natural use of speech commands to access device functionality and to retrieve processed information in real time. Furthermore, by performing onboard speech processing, the proposed technology alleviates many societal issues and challenges surrounding privacy and data protection with cloud-based solutions.This Small Business Innovation Research Phase I project aims to develop an always-on, touch-free, embedded voice trigger with ultra-low power consumption for deployment in smart products and IoT systems. This novel speech processing technology transforms analog speech signals into biologically-inspired data-efficient binary pulse trains and processes them using adaptive state models. These models can be deployed in a network of reprogrammable automata with high accuracy using only the timing information of the digital pulses produced by the user's voice. Compared with conventional speech features, pulse trains, as used in the human auditory system, are highly robust to noise and lead to extremely small footprint hardware implementations, which are ideally suited for low power on-chip processing. Furthermore, by processing only ones and zeros with automata, the proposed approach requires no arithmetic, enabling real-time, complex voice processing capability that would otherwise be achieved at prohibitive costs or be relegated to the cloud. This keyword spotting technology is pre-configurable, customized for each user, and can be used in cellphones or similar portable devices with no network connection required after calibration. It offers increased security and privacy to access portable device features by eliminating the need to send potentially sensitive data continuously online.