Decoded graph optimization method and device and storage medium

The invention provides a decoding graph optimization method and device and a storage medium, and relates to the speech recognition technology. The method comprises the steps of training an acoustic model and constructing a language model, generating an initial decoding graph according to the trained acoustic model and language model, adjusting the weight of the initial decoding graph to maximize the target function value, obtaining an optimized decoding graph, and representing the occurrence probability of the annotation word sequence corresponding to the preset voice signal under the conditions of the acoustic model and the initial decoding graph by the target function. According to the invention, a decoded picture with excellent performance can be obtained. 本申请提供一种解码图优化方法、装置及存储介质，涉及语音识别技术。此方法包括：训练声学模型和构建语言模型；根据训练好的声学模型和语言模型，生成初始解码图；通过调整初始解码图的权重，使得目标函数值最大，得到优化后的解码图，目标函数表征将预设语音信号在声学模型和初始解码图的条件下，与预设语音信号对应的标注词序列的发生概率。通过本申请可以得到性能较优的解码图。