Chinese Continuous Digital Recognition And Research In Noisy Environments

Language is the human’s most important communication tools, it can transmitinformation accurately, efficiently, and conveniently. With the continuous development ofsociety, More and more machines to participate in the activities of the human. Therefore,the relationship between humans and machines become close and important. The voiceinteraction processing has become one of the important means of human-computerinteraction.Ambient noise must be in the voice interaction as well as the identification process,currently speech recognition system for speech recognition can achieve ideal recognitionrate in quiet environment, but the recognition rate is not good in nosiy environments. so itneeds strengthen the ability of the anti-noise in the speech recognition process., it alsoneeds further study and solve problems.The paper first introduces the development status of the voice recognition technologyat home and abroad, which analyses the difficulties of speech recognition in application andneed improve continuous digital speech recognition.At present, speech recognition uses time domain and frequency domain methods. First,the paper improves the existing endpoint detection in the time domain processing section,and reset threshold to increase voice detection effection. Second, it researches the HiddenMarkov Models(HMM) in speech recognition and discusses estimate、decoding、training ofthe three basic issues and solutions around the Hidden Markov Models, gives the HiddenMarkov Models their own shortcomings and limitations. Third, it researches speechenhancement in the wavelet transform, wiener filtering, spectral subtraction and so on,compares several algorithm advantage and shortcomings. Finally, it uses matlab simulationsoftware and vs2005software giving continuous digital speech recognition in modifiedspectral subtraction, speech feature parameter extraction, improving the experimentalprogram on the hidden Markov model training and recognition. The experimental resultsshows improved spectral subtraction which can improve speech recognition rate.