【摘要】 语音识别片上系统可以实现简单的人机交互和语音控制,在家电、玩具及各种人机交互系统中有着广泛的应用前景。本文结合汉语语音特点,在TMS320VC5507芯片上实现了高性能特定人与非特定人中小词汇量孤立词识别系统。采用基于循环缓冲区的端点检测算法,双缓冲区的传输方式用于语音录制和回放,分别采用降低特征维数的DTW算法和基于连续隐含马尔可夫模型(CDHMM)的多级搜索算法作为核心识别算法,并给出实验结果。更多还原

【Abstract】 Speech recognition systems on chip (SoC) can realize simple man-machine communication and speech control, and have a bright application prospect in household appliances and various man-machine communication systems. Combined with characteristics of Chinese speech language, this paper presents a speaker dependent and independent medium and small vocabulary isolated words recognition system based on TMS320VC5507. Voice Activity Detection (VAD) algorithm based on circular buffer is implemented and dual-buffer transportation mode is used in speech recording and playing. DTW with low-dimension feature and multi-pass decoding method based on CDHMM are utilized respectively as the kernel recognition algorithms, and experiment results are reached.更多还原