Speech Recognition Of Hakka Dialect Based On Deep Learning

In recent years,due to the rapid development of big data and cloud computing as well as the popularity of artificial intelligence,using natural language for human-computer interaction-such automatic speech recognition has been a research hotspot.At the same time,speech recognition is also the primary research focus of markets such as smart homes or smart e-commerce.Although there are many speech recognition systems on the market,there are very few speech recognition and research for dialects.Under this background,we uses Hakka dialect data set as the corpus and neural network as the model structure to construct a Hakka dialect speech recognition system based on deep neural network,which has practical value for the speech recognition research of Hakka dialect.First,we introduces key technologies related to deep learning,including neural networks,Recurrent Neural Networks(RNN),Long Short Term Memory Network(LSTM),and simpler Gated Recurrent Unit(GRU).And then,we analyzed the structure and technology of the speech recognition system,and the processing technology of the speech signal is emphasized.Thirdly,through the construction of Hakka dialect corpus,we introduced the construction process of corpus in detail,including the selection,recording,collation and nuclear,voice annotation of corpus.Finally,we used TensorFlow,an open source speech recognition tool,the Hakka dialects corpus which is collected,and the current advanced recurrent neural network to complete the model training,the Hakka dialect speech recognition system is created.According to the output of dialect speech recognition test,the LSTM+CTC-based deep learning method is applied in the test data.The Hakka dialect has a good performance in speech recognition,and the accuracy of the model in a quiet environment is about 97%.