论文提出了一种连续隐Markov模型和BP神经网络相结合的、具有两次辨识过程的抗噪孤立字识别模型 .首先以连续隐Markov模型完成语音信号的时序建模并提供一次识别信息 ;以BP神经网络进行后处理 ,提取二次识别信息 ,识别结果由两次识别信息共同决定 .实验证明 ,由于有效地利用了隐Markov模型的强时序信号处理能力和BP神经网络的强模式分类和泛化性能 ,这种识别模型明显地改善了孤立字识别系统的抗噪性能 . This paper presents an anti-noise isolated word recognition model with continuous hidden Markov model and BP neural network with two recognition processes.Firstly, the sequential hidden Markov model is used to model the timing of voice signals and provide a first identification information; BP neural network, the secondary recognition information is extracted and the recognition result is jointly determined by the two recognition information.Experiments show that due to the effective utilization of strong timing signal processing ability of hidden Markov model and strong pattern classification and generalization of BP neural network, This recognition model significantly improves the noise immunity of isolated word recognition systems.