SBIR Phase I: Low-Cost Automated Speaker Verification Using Gamma Filters

这个小企业创新研究（SBIR）第一阶段的项目涉及自动说话人验证（ASV）。说话人验证是确定给定的语音样本是否属于特定的人的过程。模拟互补金属氧化物硅（CMOS）语音识别芯片目前正在开发中，它结合了集成硅耳蜗和基于听觉系统建模的神经网络。ASV将在这一技术基础上进行扩展。ASV是一项计算密集型任务，但可以通过大规模并行策略有效地解决。第一阶段将开发一种算法，该算法使用已开发的前端耳蜗处理器，结合聚焦伽玛神经网络（FGNN）和一种称为局部感受野伽玛神经网络的新结构。每个伽马滤波器将被修改为扬声器匹配滤波器。第二阶段将把ASV算法映射到专用硅（模拟或数字）上，以实现可靠、紧凑、低成本的ASV处理器。ASV对于低安全性的访问控制系统很有用。一种低成本、紧凑的ASV技术将作为人机界面的组成部分应用于现有的消费电子产品中。在家里，应答机、长途电话和其他低安全性的设备将受到保护，只有特定的扬声器才能进入；特定的有线电视频道将受到家长控制的保护；会说话的玩具会被编程，让它们认识自己的“主人”。处于睡眠或屏幕保护模式的电脑可以通过授权的声音激活。许多其他潜在的消费者应用都是可能的。

This Small Business Innovation Research (SBIR) Phase I project addresses automated speaker verification (ASV). Speaker verification is the process of determining whether a given voice sample belongs to a particular person. Analog complementary metal-oxide-silicon (CMOS) speech-recognition chips are presently under development, combining integrated silicon cochleae and neural networks based on auditory system modeling. ASV will expand on this technology base. ASV is a computationally intensive task but can be solved efficiently with massively parallel strategies Phase I will develop an algorithm that uses a developed front-end cochlear processor in conjunction with a focused gamma neural network (FGNN) and a novel structure called a local receptive field gamma neural network. Each gamma filter will be modified to behave as a speaker-matched filter. Phase II would map an ASV algorithm onto dedicated silicon (analog or digital) to realize a reliable, compact, low-cost ASV processor. ASV is useful for low-security access-control systems. A low-cost, compact ASV technology will find applications in existing consumer electronics as a component of the human-machine interface. In the home, answering machines, telephone long-distance access, and other low-security devices will be protected so that only particular speakers have access; specific cable-TV channels will be safeguarded by parental control; and speaker-trained toys will be programmed to know their `owner.` Computers in sleep or screen-saver mode could be activated by an authorized voice. Many other potential consumer applications are possible.