Research on the Architecture of the Speech Recognition System and the Computer Aided Design System for Signal Processing LSIs.

为了实现高性能的语音识别系统，需要开发实时引用大量知识库的技术。在语音识别的实际应用领域中，不仅要求高识别率，而且要求低成本。考虑到这两种需求，本研究旨在研究一种适用于语音识别和LSI架构设计系统的并行处理架构，特别是用于生产数字信号处理LSI。在并行处理方面，开发了一个专门的实验系统，它是由具有通信链路的专用微型计算机组成的多处理机系统。语音识别中的多种模式匹配过程可以通过交换中间匹配而并行进行。通过页面词汇识别任务对系统进行了测试。实验结果表明，并行处理能有效地结合多种词识别范式。本研究的第二个目标是为lsi构建一个架构设计系统。该系统的输入是用专门的高级规范语言描述的对所需要功能的规范描述，输出是寄存器传输级的硬件描述。通过使用传统的CAD系统，可以从这些描述生成lsi。这项研究展示了将人工智能技术应用于设计问题的一种新的可能性。

In order to realize a high performance speech recognition system, techniques to refer large amounts of knowledge base in real time should be developed. In the actual application areas of speech recognition, not only high recognition rate but low cost is strongly requested. Considering both requirements, this research aimed to study a parallel processing architecture suitable for the speech recognition and LSI architecture design system which particularly intended to produce digital signal processing LSIs. As for the parallel processing, a special experimental system was developed as a multiprocessor system composed of Transputers which are special microcomputers with communication links. Many kinds of pattern matching processes in speech recognition can be performed concurrently by exchanging intermediate matching. The system is tested by the task of parge vocabulary words recognition. It was verified that the parallel processing could effectively combine several kinds of paradigms developed for the word recognition.The second objective of this research is to construct an architecture design system for LSIs. The input to this system is a specification description on the required function which is described by special high level specification language and the output is the hardware description at the register transfer level. LSIs can be produced from these decription by using the conventional CAD system. This study sdhowed a new possibility applying artificial intelligence technique to design problems.