基于语音合成的机器翻译机器人设计

The Design of a Machine Translation Robot Based on Speech Synthesis

针对传统翻译机器人在嘈杂环境下无法准确识别语音,导致机器翻译效果不佳的问题,提出设计一个基于语音合成的机器翻译机器人。首先,采用翻译机器人语音系统中的语音硬件设备进行语音数据采集;然后通过语音信号预处理方法分别进行数据预加重、分帧加窗和端点检测;之后利用语言模型进行语音信号特征提取;最后对翻译机器人语音进行编码处理,以实现精准识别。实验结果表明,在采样频率为800 Hz的声卡录音实验中,改进的双门限算法进行端点检测的所用时间仅为0.17 s,识别率高达96.78%,对比于现有的双门限算法的所用时间低了4.78 s,识别率高出了33.36%。在语音识别系统测试中,本系统对合成语音的识别率保持在90%及以上,识别率较高。由此说明,本系统进行语音识别的所用时间更短,检测正确率更高,为翻译机器人语音交互提供了有效的数据支撑。

In view of the problem that the traditional translation robot cannot accurately recognize the speech in a noisy environment,which leads to the poor effect of machine translation,it is proposed to design a machine translation robot based on speech synthesis.Firstly,the speech hardware equipment in the translation robot speech system is used to collect data enhancement,frame window and endpoint detection through the speech signal pretreatment method;then,the language model is used to extract speech signal features;finally,the translation robot speech is coded to realize accurate recognition.The experimental results show that in the sound card recording experiment with a sampling frequency of 800 Hz,the time of the modified dual threshold algorithm for endpoint detection is only 0.17 s,and the recognition rate is as high as 96.78%,which is 4.78 s lower than the existing dual threshold algorithm,and the recognition rate is 33.36%higher.In the test of speech recognition system,the recognition rate of the system system is kept at 90%or above,and the recognition rate is high.This shows that the system takes a shorter time for speech recognition and has a higher detection accuracy,which provides effective data support for the speech interaction of the translation robot.