基于改进卷积神经网络的汉语连续语音识别方法

Chinese Continuous Speech Recognition Method Based on Improved Convolutional Neural Network

根据当前的识别需求及标准,提取语音识别基元,采用多目标的方式,提高识别的效率,布设交叉多目标识别矩阵,同时改进卷积神经网络连续语音识别模型,采用动态识别规整完成语音识别处理。测试结果表明,与传统面向健壮自动汉语连续语音识别测试组和传统基于改进多带谱减汉语连续语音识别测试组相比,所设计的改进卷积神经网络汉语连续语音识别测试组语音误识率被较好地控制在20%以下,说明在改进卷积神经网络的辅助下,语音识别效果明显改善,针对性更强,具有实际的应用价值。

According to the current recognition needs and standards,the speech recognition primitives are first extracted,and the multi-target method is adopted to improve the efficiency of recognition,and the cross-multi-target recognition matrix is arranged,and an improved convolutional neural network continuous speech recognition model is constructed at the same time,and the speech recognition processing is completed by dynamic recognition regularization.The test results show that compared with the traditional robust automatic Chinese continuous speech recognition test group and the traditional improved multi-band spectrum minus Chinese continuous speech recognition test group,the speech error recognition rate of the improved convolutional neural network Chinese continuous speech recognition test group designed this time is better controlled below 20%,indicating that with the assistance of improved convolutional neural network,the speech recognition effect is significantly improved,more pertinent,and has practical application value.