Application Of Henan’s Dialect Speech Recognition Based On Acoustic Model

Speech recognition has a history of nearly 70 years.In these years of exploration and practice,speech recognition technologies at home and abroad have developed more mature,and they are more and more commonly used in daily life,such as Apple’s Siri and We Chat’s "Voice-to-text" module and so on,provide people with a lot of convenience.After years of development,especially after the emergence of smart phones,the speech recognition of Mandarin Chinese has been quite accurate.But today,when speech recognition has been widely used,there are still some problems that are difficult to break through.Among them,dialect recognition is one of the difficulties.Language is the carrier of culture.China has a vast territory and many dialects.It is urgent to protect dialects.Located in the Central Plains,Henan Province is a key transportation hub in my country.It has frequent exchanges with other provinces;it has a large population and ranks third among all provinces in the country.In most areas of Henan Province,whether in rural or urban areas,the use of dialects in daily communication among Henan people is very high.As a local dialect,Henan dialect is relatively close to Mandarin,but it has its own distinctive features.In order to build a Henan dialect speech recognition system,before the experiment,the research content is mainly divided into two aspects.One is the learning of the speech recognition process,and the appropriate model is selected according to the characteristics of Henan dialect;the second is a summary of the language aspects of Henan dialect.The consonants and vowels,tones,consonants,common words of Henan dialect are summarized and arranged,and a dictionary corresponding to Henan dialect and Mandarin is created for subsequent experiments.At the same time,collect as much Henan dialect corpus as possible,sort and label the collected corpus,and supplement the Henan dialect dictionary while labeling.During the experiment,in the downloaded voices of 17 cities in Henan Province,taking into account the differences in dialects between Henan regions,if they are directly integrated,they will be brought into the voice recognition model,so that the model can extract acoustic features It may be confused and unable to find a suitable rule,which will affect the recognition rate.Since Zhengzhou,Kaifeng,and Nanyang have the majority of voices,these three places are used as a typical example,and a dialect category recognition test was done.The distinguished dialects were separated for speech recognition,and the weak dialects were merged.Then perform voice recognition.Experiments show that Zhengzhou dialect and Kaifeng dialect can be classified into one category,and Nanyang dialect needs to be tested separately.In the test of speech recognition,the acoustic model uses the current mainstream LSTM model for speech recognition,which performs better in processing speech context relations and can enhance the speech recognition effect of Henan dialect;the language model uses N-gram ternary language Model,this model is more efficient in the process of converting Pinyin to Chinese characters.In addition to the three places mentioned above,because the other 14 cities have not many corpora,the data of the 17 cities are divided into the southern and northern categories in Henan Province.The northern corpus is richer and the recognition accuracy rate is average.Greater than in the south,the recognition effect of words,words,and sentences all decrease in order.The research on Henan dialect speech recognition can not only promote the development of Henan speech technology,strengthen the communication between Henan dialect and computers,but also contribute to the inheritance of Henan dialect culture.