Font Size: a A A

Research And Implementation Of Voiceprint Recognition System Based On Deep Learning

Posted on:2021-04-24Degree:MasterType:Thesis
Country:ChinaCandidate:B LiangFull Text:PDF
GTID:2428330626962661Subject:Software engineering
Abstract/Summary:PDF Full Text Request
With the advancement of technology and the rapid development of the Internet,voiceprint recognition technology plays an increasingly important role in the interaction between people and electronic products.At present,many researchers are looking at the research field of voiceprint recognition.The development of voiceprint recognition technology is becoming more and more mature,with a solid theoretical system and a high community popularity.The traditional GMM(Gaussian mixture model)is a very classic and important model in the field of voiceprint recognition,but due to its ability to model large speech data and poor processing of noise,researchers began to study sound based on deep learning Pattern recognition,the introduction of convolutional neural network(CNN)for model training.The residual network is one of the convolutional neural networks.By using the residual network to train the model,you can get good results.Based on deep learning technology,this paper studies and designs a voiceprint recognition system based on deep learning,which can effectively recognize specific speakers.The construction of the system is based on high-level languages such as Python and is based on deep learning frameworks such as Tensorflow and Keras.The speaker recognition system based on deep learning proposed in this paper mainly includes two steps: voiceprint registration stage,data training stage and speaker online recognition stage.In the voiceprint registration phase,the user's voice is collected,and then in the data training phase,the pre-emphasis,framing,and windowing algorithms are first used to preprocess the voice file and calculate the spectrogram;then,use the improved residual The neural network trains the input feature data to extract feature vectors;finally,after the extracting phase is completed,the feature map is input into the NetVLAD layer for clustering,and then the AM-Softmax loss function is used for evaluation,and then the model is saved.In the voiceprint recognition stage,the system preprocesses the collected test speech and inputs it into the trained model for classification and recognition.In addition,the system also includes the following core modules: user login registration,voiceprint registration,model training,voiceprint recognition and other modules,which can finally achieve the function of effectively completing the user's remote network for voiceprint recognition needs under the B / S architecture.Through theoretical analysis and test experiments,it can be found that compared with traditional systems,the voiceprint recognition system designed in this paper uses the network structure of the network for model training and model recognition.Compared with other convolutional neural networks,the system proposed in this paper has more High accuracy,and the system uses B / S architecture,users can operate on the browser side,which can meet the needs of users for remote voiceprint recognition.
Keywords/Search Tags:Voiceprint recognition, Deep Learning, Residual Network, Spectrogram, NetVLAD, AM-Softmax
PDF Full Text Request
Related items