Research On Speaker Gender Feature Recognition Based On Deep Learning

Gender recognition is an important field of speech recognition,which is based on the speaker's acoustic characteristics.It is one of the earliest research topic in the field of speaker recognition,sex identification of the voice signal first,and then according to the speaker recognition of men's and women's categories,respectively,can significantly reduce the search space and time,in improving the speaker identification accuracy,improve the processing speed of the system has great significance.Moreover,gender-based emotion recognition technology has always been a way for researchers to deal with emotional calculation.The traditional speaker gender recognition systems,in clean speech environment to phonemes or unit as gender recognition training corpus recognition rate is higher,in view of the telephone voice gender identification can achieve good effect;The recognition rate in continuous speech or short speech segment is greatly reduced.In the speaker's gender identification system,the selection of feature parameters has a significant impact on the recognition of the system,and the single speech feature has limitations on the improvement of recognition rate.This paper attempts to improve the recognition rate of continuous speech by using a variety of feature combinations.Firstly,the influence of the characteristics of Merle's spectral coefficients on the recognition rate is compared.Secondly,the characteristic coefficient with high recognition rate is selected as the combination feature to further improve the recognition rate of the system.In addition,due to the insufficient processing capacity of high dimensional data due to the shallow classification model,it is difficult to extract the deep characteristic information and easily fall into the local optimal solution.Therefore,the method of deep learning is introduced to classify and identify,which makes the performance of deep neural network improved.Finally,through the experimental results show that: in this paper,the speaker gender identification method to be used in the two groups of library under the recognition rate can reach more than 96%,and compared with the classification model of shallow,it to talk other recognition rate is higher,humanity has better recognition performance.