Research And Application Of Speech Recognition In Power System

Speech recognition is an important field of artificial intelligence research,which has important theoretical significance and practical value.Chinese is the most widely used language in the world.Chinese speech recognition technology has a very broad application prospects.In this paper,three modules of Chinese speech recognition system based on Gaussian Mixture Model/Hidden Markov Model(GMM-HMM)are studied and implemented,two of which are improved,and the improved GMM-HMM is applied to power system command dispatching speech recognition.Firstly,the framework of Chinese speech recognition system based on GMM-HMM is studied,and the acoustic model is implemented and analyzed.Modern Chinese is different from the Indo-European language in that it has a unique attribute of tone.The same Chinese character represents different meanings in different tones.In view of this phenomenon,this paper proposes to increase the tone model in the acoustic model.Experiments show that the tonal model can effectively distinguish the approximant characters with the same initial but different tones.Then,this paper studies the direction of continuous speech recognition.In modern Chinese,the same pronunciation may represent completely different Chinese characters.There are some differences between the recognition results of homophone and the actual semantics.In view of this phenomenon,this paper proposes to change the monophone model to the context-based triphone model.According to the principle that the presence of the current word is related to the appearance of the previous word,the training set is statistically calculated word transition probability,and the word transition probability is added to the language model.Experiments show that the triphone model and word transition probability can effectively improve the recognition rate of homophones.Finally,the speech recognition system based on the improved model is implemented,and the system is applied to the power voice dispatching system to record dispatching instructions automatically.Experiments show that the speech recognition system after the improved model can be applied to Chinese speech recognition effectively.