小波神经网络(WNN)具有高度的非线性映射功能及强大的自适应能力，但是WNN算法存在易陷入局部极小值，收敛速度慢。而人工蜂群算法(ABC)具有很强的全局搜索能力及较快的收敛速度。两者优势互补，已结合应用于语音识别中。本文对ABC算法做出改进，在采蜜蜂和观察蜂阶段各提出一个新的解搜索方程，采取自适应的双搜索方式(Adaptive Double Search)求解，从而提高算法的收敛速度和收敛精度。并将其和WNN算法进行结合，组成一种训练神经网络的新算法ADSABC-WNN，该算法既能克服WNN算法的缺点，又能保存双方的优点。实验结果表明，与传统ABC算法优化小波神经网络相比，识别率提高均有所提高，其中在词汇量为50时识别率提高了4.51%。将实验结果与其他方法优化的小波神经网络模型进行比较，在噪声环境下，该混合模型可以有效地减少识别时间，而且可以明显提高网络的训练速度和语音识别的识别率。

Wavelet neural network(WNN) has a highly nonlinear mapping function and a strong adaptive ability, but the WNN algorithm tends to fall into local minimum and converges slowly. The artificial bee colony algorithm (ABC) has a strong global search ability and faster convergence rate. The two algorithms complement each other and have been applied to voice recognition. In this paper, ABC algorithm is improved. A new solution search equation is proposed in the bee mining and the observation bee. Adaptive Double Search is used to solve the problem, so as to improve the convergence speed and convergence accuracy. And we combined with WNN algorithm to form a new algorithm ADSABC-WNN which can not only overcome the shortcomings of WNN algorithm but also save the advantages of both. The experimental results show that compared with the traditional ABC algorithm, the recognition rate improves, and the recognition rate increases by 4.51% when the vocabulary size is 50. Compared with the wavelet neural network model optimized by other methods, this hybrid model can effectively reduce the recognition time under the noise environment, and can obviously improve the training speed of network and the recognition rate of speech recognition.