基于吴语和普通话混合的无差别语音识别

Undifferentiated speech recognition based on the mixture of Wu and Mandarin

针对使用方言的人数多,方言的识别效果较差问题,本文以上肢康复外骨骼的语音指令作为识别对象,使用Python的第三方库Keras和Librosa建立了一种DNN-HMM[深层神经网络(Deep Neural Network),隐马尔可夫模型(Hidden Markov Model)]模型,对吴语方言与普通话混合的语音样本进行无差别语音识别研究和实验。结果表明,本文研究的模型对于吴语方言与普通话混合的语音样本识别率能够达到81%,具有较高的识别精度;且对于单独的吴语方言识别率有65%左右。该研究结果可以进一步的应用于其他地区方言的语音识别,为智能康复外骨骼的控制研究和推广提供了理论和实验基础。

In view of the large number of people use dialects and the poor recognition effect of dialects,the voice instructions of the upper limb rehabilitation exoskeleton are used as the recognition object in this paper,where a DNN-HMM(Deep Neural Network and Hidden Markov Model)model is established using the third-party library Keras and Librosa of Python to carry out an undifferentiated speech recognition experiment on the mixed voice samples of Wu dialect and Mandarin.The experimental results show that the recognition rate of the model can reach 81% for the mixed speech samples of Wu dialect and Putonghua,and it has a high recognition accuracy.The recognition rate of Wu language alone is about 65%.The research results can be further applied to speech recognition in other dialect areas,providing a theoretical and experimental basis for the control research and promotion of intelligent rehabilitation exoskeleton.