基于DSP嵌入式英语识别系统开发研究

为了适应我国互联技术与移动设备的快速发展,非常迫切需要设计一款不但又友好,而且能方便用户操作的新系统。文章使用两级识别网络,立足于连续隐含马尔可夫模型作为网络架构,构建一个每秒钟内能够处理的百万级的机器语言指令数的16位定点DSP语音芯片平台,设计出一款嵌入式英语实时识别系统;最后通过相关实验表明:该系统能够提升英语识别性能,同时还能使得内存占用量与计算复杂度得到降低。

Research and development of embedded English recognition system based on DSP

in order to adapt to the rapid development of Internet technology and mobile devices in China,it is very urgent to design a new system which is not only friendly,but also convenient for users to operate.Based on the continuous hidden Markov model,a 16 bit fixed-point DSP speech chip platform with millions of machine language instructions per second is constructed by using a two-level recognition network.An embedded real-time English recognition system is designed.Finally,the relevant experiments showed that the system can improve the performance of English recognition and make the Memory usage and computational complexity to be reduced。