Research And Implementation Of Speech Recognition Algorithm For Patients With Aphasia

Aphasia is a common disease in daily life.The rehabilitation treatment of aphasia patients has attracted more and more attention.Due to the high cost of treatment and the shortage of rehabilitation doctors,a large number of aphasia patients failed to receive effective training in the later stage.In order to improve the difficulty of training,it is necessary to use intelligent aphasia rehabilitation training.The core technology of intelligent system is speech recognition.Research and implementation of speech recognition algorithm for aphasia patients.Firstly,six phonetic types of aphasia patients are analyzed,and a corpus of aphasia patients is constructed.The process of constructing corpus includes speech signal preprocessing,signal denoising,feature extraction,speech de duplication,speech segmentation and speech annotation.After analyzing and comparing the common denoising algorithms,Kalman filter denoising algorithm is used to denoise the collected speech signal.After studying and comparing the common feature extraction algorithms,Mel frequency cepstrum coefficient is used for speech feature extraction.Finally,the construction of the corpus is completed.Based on the establishment of aphasia patient corpus,speech recognition is carried out for patients.Hidden Markov model(HMM)is the mainstream algorithm in the field of traditional speech recognition.On this basis,the forward backward algorithm,Viterbi algorithm and Baum Welch algorithm in the model are reproduced,studied and analyzed.It is found that Baum Welch algorithm has the disadvantages of slow convergence speed and low recognition rate in the model training.K-means algorithm is a clustering algorithm.The algorithm is fast and simple.It has high efficiency for large data sets and is scalable.It can be repeated until a certain termination condition is met.On this basis,the Baum Welch algorithm in HMM model is improved by K-means algorithm.The experimental data show that the algorithm can jump out of the local limit,search for the global optimal value,shorten the speech training time,and improve the average recognition rate from 84.2% to 88%.