基于改进ABC与Attention-CTC的语音识别技术研究  

Speech Recognition Based on Improved ABC with Attention-CTC

作　　者：张竞[1] ZHANG Jing(BeiJing Institute of Economics andManagement,Beijing 100102,China)

机构地区：[1]北京经济管理职业学院,北京100102

出　　处：《自动化与仪器仪表》2025年第2期14-17,23,共5页Automation & Instrumentation

基　　金：北京市教育科学规划项目《协同理论视域下高职产业学院建设模式与路径研究》(CGDB21211)。

摘　　要：现阶段的智能化科学技术对于人类听觉系统的分析理解还很弱,无法将其应用于英语语音的识别当中。因此,研究针对智能化英语语音识别遭遇的困难与挑战,将改进后的人工蜂群算法、注意力机制与联接主义时序分类算法相融合,创新性地提出了一种基于改进人工蜂群算法与联接主义时序分类算法的语音识别模型。实验结果表明,研究所提模型的英语语音识别准确率达到了96.23%,单词错误率和字符错误率分别仅为4.67%与1.98%。且研究提出的新型英语语言识别模型P值最高为95.46%,R值最高为92.29%,F1值最高为93.84%,平均检测时间最短仅为2.54 s。由此可知,研究所提新型语音识别模型具有不错的语音特征提取与识别能力,能为智能化英语语音识别提供一定程度的理论支持。At the present stage,intelligent science and technology is still weak in analysing and understanding the human audito-ry system,and cannot be applied to the recognition of English speech.Therefore,to address the difficulties and challenges encoun-tered in intelligent English speech recognition,the study integrates the improved artificial bee colony algorithm,attention mechanism and connectionist temporal classification algorithm,and innovatively proposes a speech recognition model based on the improved artifi-cial bee colony algorithm and connectionist temporal classification algorithm.The experimental results show that the English speech recognition accuracy of the proposed model reaches 96.23%,and the word error rate and character error rate are only 6.67%and 2.08%,respectively.The proposed novel English speech recognition model has the highest P-value of 95.46%,the highest R-value of 92.29%,the highest F1-value of 93.84%,and the shortest average detection time of only 2.54 seconds.It can be seen that the new speech recognition model proposed in the study has good speech feature extraction and recognition ability,which can provide a certain degree of theoretical support for intelligent English speech recognition.

关 键 词：ABC 注意力机制 CTC 英语 语音识别 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]