Speech Recognition Based On The Intelligent Mandarin Language Healing System Of Deaf Children

Our society has been paying more and more attention to the Language Healing System of Deaf Children today. With the popularization of network and multi-media computer, the demand of language healing system software is increasing, and the requirement for its intelligence also becomes higher and higher. This paper comes from the development of the Intelligent Mandarin Language Healing System of Deaf Children. The Deaf Children Language Healing System is different from the existing speech recognition products. According to the pronunciation defect of deaf children, it processes the parameters (pitch, meaning ,emotion) respectively, Then the deaf children can have a clear vision of their own speech from the cartoon and the visual graph which are transformed from the speech parameters. Because the cartoon and the visual graph are simple and easy to understand, the children also can understand the physical meaning reflecting on the phonation organs through vision.According to the speech training process and the application environment, this paper introduces the preprocessing of speech briefly and mainly studied the isolated mandarin syllables pitch recognition and speech recognition. Effective recognition methodes which are suitable for the Language Healing System of Deaf Children are put forward in this paper from the features of modern Chinese.In the preprocessing part, this paper focuses on the research of endpoint detection. For a quite long time, the endpoint detection algorithm is according to the time-domain characteristics of speech, with the development of speech processing, A lot of new endpoint detection methods based on frequency domain appeared. Each method has its advantages and disadvantages. We should choose the suitable method according to the application environment. There is less background noise in a quiet room which is used for the Deaf Children Language Healing System. The experiment showed that the endpoint detection method based on short-time energy and short-time average zero-crossing rate could meet the need of the system and this method is simple, and easier to real-time realized. On the basis of original algorithm, this paper detect three endpoints of a monosyllable which are used for the latter processing. These three points are the starting point of the whole syllable, the starting point of the sonant part and the end point of the whole syllable. Finally, the speech enhancements are introduced briefly which will be used for a noisy environment.Pitch recognition is an important part of the Language Healing System of Deaf Children. The speed of speaking can be affected by emotional, stylistic and environmental factors and it may change from time to time. As a result, the length may vary greatly even when pronouncing the same word. This uncertainty of length and personal differences make the pitch recognition difficult. This paper presents an efficient algorithm for Tone Recognition of Isolated Mandarin Syllables. In this paper, four parameters that are used for pitch recognition latter are obtained by curve fitting of the first degree. In addition, a pre-classification method was adapted used for improving the third tone recognizing rate tone. The experimental results show that by using these new methods, an accuracy rate of 97.84% can be achieved in Mandarin tone recognition without prior training.Speech recognition is another important part of the Language Healing System of Deaf Children which is related to the children's comprehension of speech. This paper presents an efficient algorithm for speech recognition of isolated mandarin syllables by recognizing the unvoiced part and voiced part separately. The Fourier coefficients are used in the recognition of unvoiced part which can make the deaf children experience the pronunciation better. In the voiced part recognition, the formant and the short-time energy are adapted because of its close connection to speech organs. Then the system can present pronouncing suggestion according parameter comparison result.In a word, the algorithms presented in this paper which are based on the Chinese speech characteristics is suitable for the Language Healing System of Deaf Children, and an satisfactory accuracy rate can be achieved in the recognition.