面向政务系统的大数据语音识别系统应用及研究  

Application and Research of Big Data Speech Recognition System for Government Affairs System

作　　者：夏美艺 范灵 牛青松 桂鹂娟 Xia Meiyi;Fan Ling;Niu Qingsong;Gui lijuan(Qinghai Communications Technical College,810003;Qinghai Provincial Institute of Science and Technology Information Co.,Ltd,810007)

机构地区：[1]青海交通职业技术学院,青海西宁810003 [2]青海省科学技术信息研究所有限公司,青海西宁810007

出　　处：《现代科学仪器》2024年第1期155-160,共6页Modern Scientific Instruments

基　　金：青海省2021年重点研发与转化计划项目《政务智能语音识别系统研究与应用》,项目编号:2021-GX-116。

摘　　要：当前对于大数据语音识别系统在政务系统应用中存在诸多缺陷,因此,研究将LSTM与CTC进行融合得到了LSTM-CTC声学模型,并进一步优化得到BiLSTM-CTC声学模型,同时验证其有效性。实验结果表明,在训练轮数为8时BiLSTM-CTC模型的WER值为60.38%,在训练轮数为16时,BiLSTM-CTC声学模型的WER值为11.87%,均低于对比模型。同时,在实际的政务系统大数据语音识别中,BiLSTM-CTC声学模型在安静与低噪声环境下均具有较高的识别准确性,平均识别率分别为92.6%和85%。综合来看,BiLSTM-CTC声学模型在识别政务系统的大数据语音中具备较高的准确性,在实际中可以有效推进政务系统语音识别功能的发展。The development of artificial intelligence and the advancement of intelligent government have put forward higher requirements for the timeliness of information processing in government systems.However,the current application of big data speech recognition systems in government systems is not mature enough,and there are many defects.Based on this,the research fused LSTM and CTC to obtain the LSTM-CTC acoustic model,and further optimized to obtain the BiLSTM-CTC acoustic model,while verifying its effectiveness.The experimental results show that the WER value of the BiLSTM-CTC model is 60.38%when the number of training rounds is 8,When the number of training rounds is 16,the WER value of the BiLSTM-CTC acoustic model is 11.87%,which is lower than the comparison model.At the same time,in actual government system big data speech recognition,the BiLSTM-CTC acoustic model has high recognition accuracy in both quiet and low noise environments,with an average recognition rate of 92.6%and 85%,respectively.In summary,the BiLSTM-CTC acoustic model has a high accuracy in identifying big data speech in government systems,and is of great significance in promoting the speech recognition function of actual government systems.

关 键 词：政务系统 大数据 语音识别系统 声学模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]