Study On Speech Control Of Turning Movements Of The Nursing Bed

It is research hotspot in recent years to control the device control by speech recognition, and it has a wide range of application. By speech recognition the computer can understand human language with a convenient, natural and fast way. Although the technology of speech recognition has made tremendous progress, it still has a shortage of recognition rate and real time.The speech control mode of turning movements of the nursing bed uses speech recognition technology to achieve speech control of turning movements of the bed. On the basis of analyzing the process of speech recognition, this paper is in-depth studies of using the fuzzy clustering algorithm to obtain voice signal endpoint detection threshold automatically, and uses Principal Component Analysis (PCA) method to reduce the dimension of the speech signal data set, so that the training time of Hidden Markov Model (HMM) speech recognition reduces significantly. This work is supported by the Technical Plan Project of Guangdong Province of China whose name is Device with Multifunctional Caring and Physiological Parameters Remote Monitoring and project code is2010A030500006. Additionaly, this work is supported by the Special Funds for Technology Development Projects of Foshan City of Guangdong Province of China whose name is Multifunctional Nursing Robot and project code is FZ2010013.First, this paper introduces the content-related background and significance of the study, analyzes the status of the motion control of the nursing bed, the status of speech recognition study, the status of voice activity detection and the status of speech feature dimensionality reduction.Second, the characteristic data of voice signals are pretreated, including pre-emphasis of voice signals and endpoint detection of voice signals. As environment noise impacts endpoint detection of speech recognition, on the basis of the analysis of a variety of endpoint detection methods, this paper adopts fuzzy clustering algorithm to calculate the appropriate endpoint detection threshold to match with the background noise to improve the accuracy of endpoint detection. Then extract Mel Frequency Cepstral Coefficients (MFCC) feature parameters from endpoint detected voice to provide a number of valid speech signal feature data.Third, analyze the principle of Dynamic Time Warping (DTW) and HMM speech recognition algorithm, according to the user of motion control system of nursing bed by speech control is non-specific, the paper selects HMM as speech recognition algorithm of speech control of turning movements of the nursing bed.Fourth, it exists a problem of MFCC speech features data redundancy and high dimension which leads to large training of time, so this paper has designed HMM speech recognition algorithm based on PCA to reduce dimensions of MFCC speech characteristics, retain the main ingredient of MFCC speech characteristic, then use HMM algorithm for speech training and recognition so as to achieve the purpose of improving speech recognition real-time under the condition of remain speech recognition rate.Fifth, design a motion control system of the nursing bed, develop the speech control system of motion control of the nursing bed by using Visual C#and Matlab mixed programming.Sixth, experiments of speech recognition algorithm designed are done and the speech control system of turning movements of the nursing bed is debugging. Experimental results show that:(1) The method of endpoint detection based fuzzy clustering algorithm obviously improves the accuracy of speech endpoint detection in the case of existing background noise.(2) Dimensionality reduction by PCA effectively reduce speech training time under condition of ensuring the speech recognition rate, so as to improve speech recognition real-time, and achieve good recognition effect.Finally, summarizes the work carried out on this paper, and gives the recommendations for further research.