基于OpenRISC1200的孤立词识别系统设计与实现

Design & Implementation of Isolated Speech Word Recognition System Based on OpenRISC1200

介绍了一个基于32位OpenR ISC1200开放源码微处理器内核的小词汇量孤立词语音识别系统结构。根据软硬件协同设计方法,研究和比较了孤立词语音识别各个环节的计算量,合理分配软硬件资源,并提出一种适合FPGA(现场可编程门阵列)实现的动态时间规正硬件实现思路,大大缩短识别响应时间。该系统在成本和知识产权方面都较市场上流行的ARM、8051等内核有优势。实验结果表明,在特定场合下,该系统对于100个词组的平均识别响应时间少于2 s,特定人识别率95%以上,非特定人识别率87%以上。

Embedded speech recognition has satisfactory prospect in consumer electronics and intelligent control fields abroad. This paper describes a small vocabulary isolated speech word recognition system architecture based on 32-bit OpenRISC1200 open source embedded microprocessor. According to software/hardware collaborated design method, this paper studies and compares the computing load in the speech recognition flows, distributes the software and hardware resource reasonably, and suggests a method of computation with DTW suitable for realization with FPGA. It significantly reduces the response time. This system has some advantages in cost and intelligence properly compared with ARM and 8051. The experiment results show that it recognizes up to 100 phrases with an average duration less than 2 seconds, and the recognition accuracy reaches more than 95% for specific speaker and 87% for non-specific speaker.