一种改进的孤立词语音识别系统设计

The design of an improved isolated word speech recognition system

针对孤立词语音识别系统设计一个改进的系统。该系统通过维纳滤波滤除噪声得到估计语音,对该语音进行双门限端点检测和特征提取得到端点范围内的特征向量,采用改进动态时间规划算法计算该特征向量与模板特征向量之间的欧式距离得到识别结果。仿真对比实验结果表明,改进系统在识别效果和识别效率方面有提高。

An improved system is proposed for the isolated word speech recognition system. In this improved system, wiener filtering is used to filter out noise effectively and to obtain estimation of speech. Double threshold endpoint detection and feature extraction are carried out on the speech to obtain feature vector within the scope of the endpoint. Recognition results can then be got by using the improved dynamic time programming algorithm to calculate the Euclidean distance between the vector and template feature vector. Simulation experiments show that the improved system is better in terms of recognition effect and the recognition efficiency.