基于语音识别技术的在线语言翻译交互学习系统的设计与实现

Design and implementation of online language translation interactive learning system based on speech recognition technology

为实现在线语言翻译交互学习,基于语音识别技术设计并实现了一套在线语言翻译交互学习系统。系统通过采用CNN结合CTC模型与BiLSTM模型构建语音识别声学模型,并采用Transformer模型作为语言模型,实现了语音准确识别,识别字错率低于16%,句识别准确率超过95%;然后基于FastSpeech2x声学模型和MelGAN声码器模型,对系统语音合成算法进行设计,实现了高质量的语音合成,MOS评分超过4分。最后,通过对系统语音采集、语音识别和语音合成各个功能模块进行测试,验证了系统性能。测试结果表明,系统各个模块功能完善,可正确执行语音采集、语音识别和语音合成等操作,实现高效率的语音采集、高准确率的语音识别,以及高质量的语音合成,且系统实现语音交互时间不超过5 s,满足语音交互时间需求,且各个功能模块完成相应功能指令的正确率均超过96%,具有良好的用户体验感,可用于实际在线语言翻译学习应用中。

In order to realize online language translation interactive learning,an online language translation interactive learning system is designed and implemented based on speech recognition technology.The system uses CNN,CTC model and BiLSTM model to build the acoustic model of speech recognition,and uses Transformer model as the language model to achieve accurate speech rec-ognition.The recognition error rate is less than 16%,and the sentence recognition accuracy rate is more than 95%.Then,based on the FastSpeed2x acoustic model and MelGAN vocoder model,the system speech synthesis algorithm is designed to achieve high-qual-ity speech synthesis,and the MOS score exceeds 4 points.Finally,the performance of the system is verified by testing the various functional modules of voice acquisition,speech recognition and speech synthesis.The test results show that each module of the system has complete functions,can correctly perform voice acquisition,speech recognition and speech synthesis,and can achieve efficient voice acquisition,high accuracy speech recognition,and high-quality speech synthesis.The system can achieve voice interaction time of no more than 5s,meet the demand of voice interaction time,and the correct rate of each functional module to complete corre-sponding functional instructions is more than 96%,It has a good sense of user experience and can be used in practical online language translation learning applications.