摘要
语音识别系统在音频质量较差时,经常出现识别错误的情况,为提高识别精度,基于连续隐马尔科夫模型设计英语翻译机器人语音识别系统。在硬件中,设计音频信号接收器和机器人音频识别模块主处理器。在软件中,对音频信号量化并预加重处理,计算帧移距离与每帧长度之间的比值,获取模拟信号转换频率与基本单位量化指标;基于连续隐马尔可夫模型构建语音文本解编码器,计算窗函数的宽度,在网格中获取马尔科夫链概率路径,比较不同概率路径的复杂度;设计英语翻译机器人语音识别算法,得到英语翻译机器人的语音识别结果。由实验数据可知:该系统在三种不同音频质量下的语音识别准确率均在75%以上,较其他系统更稳定,在同等音频质量下,准确率更高,可见连续隐马尔可夫模型的语音识别系统优于其他系统。
In order to improve the recognition accuracy,an English translation robot speech recognition system based on continuous hidden Markov model is designed.In the hardware,the audio signal receiver and the main processor of the robot audio recognition module are designed.In the software,the audio signal is quantized and pre emphasized,the ratio between the frame shift distance and the length of each frame is calculated,and the conversion frequency of analog signal and the quantization index of basic unit are obtained;Based on the continuous hidden Markov model,the speech text coder is constructed,the width of the window function is calculated,the Markov chain probability path is obtained in the grid,and the complexity of different probability paths is compared;The speech recognition algorithm of English translation robot is designed,and the speech recognition results of English translation robot are obtained.According to the experimental data,the speech recognition accuracy of the system under three different audio quality is more than 75%,which is more stable than other systems.Under the same audio quality,the accuracy is higher.It can be seen that the speech recognition system of continuous hidden Markov model is better than other systems.
作者
赵力瑾
高攀
ZHAO Lijin;GAO Pan(Yulin University,Yulin Shannxi 719000,China;Huadian Yulin Coal Technology Engineering Ltd.Co.Yulin Shaanxi 719000,China)
出处
《自动化与仪器仪表》
2022年第6期191-194,200,共5页
Automation & Instrumentation
基金
陕西省教育厅2020年度一般专项科学研究计划项目“榆林特色好产品”汉英动态术语库标准化建设与应用(20JK0463)。
