SBIR Phase I: A Noise Canceller for Enabling Speech Recognition in Noisy Environments

9861548来自信号分离技术（SST）的这个小型企业创新研究第一阶段项目应用噪声消除来使语音识别系统能够在诸如汽车的噪声环境中更有效地工作，此外，语音麦克风的免提操作是期望的能力，例如，在语音命令到汽车电话，在汽车电话上交谈，和非紧急命令操作窗口。 基于无信号参考的奇异值分解（SFR-SVD）噪声消除器是根据陆军SBIR合同生产的，这种新颖而高效的噪声消除算法现在被认为能够解决噪声环境问题。 当应用噪声消除时，将通过测量正确解释命令的百分比的改进来评估所提出的应用。语音识别系统正受到相当大的关注，这反映在商业周刊1998年的特别报告中，该报告的副标题为“语音技术是计算领域的下一件大事”，该报告讨论了语音识别和软件系统的主要制造商的进展。 这种系统在存在发动机、道路和其他噪声的情况下显著地失去可靠性。 在这种情况下，信号分离技术提供了其技术：一种高精度信号分离噪声消除器，具有在宽频率范围内有效工作的潜力，可用作语音识别系统的输入。 这项技术最初的目标市场是豪华汽车，但预计在未来将渗透到超过1亿辆的汽车市场。 仅在美国，就有1000万辆豪华汽车，据估计，在不久的将来，50%的豪华汽车可能会有语音识别器。

9861548 This Small Business Innovation Research Phase I project from Signal Separation Technologies (SST) applies noise cancellation to enable speech recognition systems to function more effectively in noisy environments such as automobiles where, in addition, hands-free operation of the speech microphone is a desired capability, e.g., in voice commands to car phones, conversing on car phones, and non-emergency commands to operate windows. Building on the Signal-Free Reference using Singular-Value Decomposition (SFR-SVD)-based noise canceller produced under an SBIR contract from Army, the novel and highly effective noise canceling algorithms are now believed to be capable of solving the noisy environment problem. The proposed application will be evaluated by measuring the improvement in percentage of correctly interpreted commands when noise canceling is applied. Speech recognition systems are receiving considerable attention as reflected in Business Week's 1998 Special Report with the subheading "Speech Technology is the Next Big Thing in Computing" which discussed progress at the major manufacturers of speech recognition and software systems. Such systems significantly lose reliability in the presence of engine, road, and other noise. In this context, Signal Separation Technologies proffers its technology: a high accuracy signal separation noise canceller which has the potential to work effectively over a wide frequency range for use as an input to speech recognition systems. The initial target market for this technology is luxury cars, but penetration into the total car market of over 100 million is anticipated in to follow. In the United States alone, there are 10 million luxury cars and it is estimated that a good 50% are likely to have speech recognizers in the near future.