Font Size: a A A

Research On Voiceprint Recognition Technology Based On Deep Learning

Posted on:2021-01-03Degree:MasterType:Thesis
Country:ChinaCandidate:L ZhangFull Text:PDF
GTID:2428330605450726Subject:Electronics and Communications Engineering
Abstract/Summary:PDF Full Text Request
With the development of deep learning,the research direction has expanded from the initial picture field to speech recognition,voiceprint recognition and natural language processing.Among them,voice recognition is regarded as a biometric authentication technology for identifying people's identity.This technology is used to extract and process the speaker's speech signal without interference.Personality information.The emergence of deep learning solves the over-fitting problem caused by local optimization in the traditional model,and can better learn the characteristics of the speaker.In order to make deep learning better in voiceprint recognition,this paper studies from the following aspects.?,The research background and historical development of voiceprint recognition are introduced.The shortcomings of traditional voiceprint recognition methods are analyzed and deep learning is introduced as the research direction of voiceprint recognition.?,The front-end processing of speech signals is studied,including the research on speech signal pre-processing and speech signal enhancement techniques.Based on the basic spectral subtraction method,an adaptive multi-window spectrum subtraction method is proposed.The algorithm adaptively adjusts the parameters according to the signal-to-noise ratio of the input speech signal,and solves the problem of important information missing caused by the fixed parameters in the basic spectral subtraction.The noise reduction effect of the voice signal.?,The voiceprint recognition method based on long-term and short-term memory network(LSTM)is studied to construct the LSTM network model.Based on the existing LSTM network model,the input sequence is segmented by sliding window to maintain the continuity of the speech signal.In the judgment of voiceprint recognition,the segment similarity loss function is introduced,which can deeply mine the local information of the speaker.?.A method based on the spectrogram-based convolutional neural network(CNN)voiceprint recognition is studied.The speech map of the speech message is input to different convolutional neural networks,and the residual network and the VGGNet network are merged to form a VGGV network,and the fully linked layer and the average pooled layer in the converged network are optimized.?,The microphone was used to collect the sound,and a voiceprint database containing 60 people(38 males and 22 females)was constructed and applied to the long-short memory network and the convolutional neural network.The dataset was compared with the standard TIMIT database.Performance under the voiceprint recognition system.Based on the front-end processing of the speech enhancement algorithm,this paper studies the application of deep learning in voiceprint recognition.Using existing network models,the network is merged and improved to make the recognition rate of the voiceprint recognition system reach 93.5%.
Keywords/Search Tags:speech enhancement, voiceprint recognition, adaptive multi-window spectrum subtraction, deep learning
PDF Full Text Request
Related items