Research On Nursing Needs Recognition Based On Incomplete Speech Of The Elderl

With the acceleration of the aging process and people’s continuous pursuit of the quality of life,the society is paying more and more attention to the needs of elderly care.The elderly need nursing staff to take care of their daily life due to illness or physical reasons.However,when the elderly speak,they will have incomplete words,substandard pronunciation,swallowing,or the voice is too small to be covered by surrounding environmental noise,etc.Thus,it is not possible for nursing staff to correctly understand the needs expressed by this incomplete voice of the elderly.To this end,this article studies the voice recognition of the elderly and the matching of incomplete voices with care needs,and recognizes their care needs from the incomplete voices of the elderly.The main work content includes:(1)Aiming at the situation that the aging of the vocal organs of the elderly leads to lower speech amplitude and greater influence of noise,the speech enhancement algorithm is emphasized.An improved speech enhancement algorithm based on signal subspace method is proposed,which introduces perceptual domain features,takes into account the simultaneous and temporal masking characteristics of the human ear,uses variance normalization to reduce residual noise,and uses cepstral domain oblique projection processing colored noise,to get an enhanced speech signal.Experiments have proved that the improved speech enhancement algorithm improves the denoising effect and increases the signal-to-noise ratio of speech.(2)The acoustic model of speech recognition is studied.Mainly studied the Gaussian Mixture Model-Hidden Markov Model and Deep Neural Network Model-Hidden Markov Model.First use the Gaussian Mixture Model-Hidden Markov Model training data,label the feature vector,and get Initial probability and state transition probability,and then replace the Gaussian mixture model with a deep neural network,and train on the marked training set.As a result,the deep neural network is used to describe the relationship between the syllable state and the feature vector.(3)The demand matching of incomplete speech recognition results for the elderly is studied.Mainly research text preprocessing and word vector training.Text preprocessing includes stop word removal,word segmentation and keyword expansion.The training of word vector includes word vector training model and semantic similarity calculation.A semantic similarity calculation method based on weighted cosine similarity is proposed.A threshold is set on the cosine similarity calculation result to screen overlapping words and non-overlapping words and assign weights.The similarity of two sentences is calculated by linear weighting.calculate.