Research On Automatic Voiceprint Recognition Technology Based On Waveform Analysis

As information technology develops,the era of Internet of Things(IoT)has arrived with an ascending amount of IOT equipment employed in our society.Among them,smart home system has played an important role.Voices-which involve semantics as well as voiceprints,can be recognized by machine learning,thus helping people use the smart home system efficiently and conveniently.In the past,voiceprint recognition could analyze speakers,but it failed to recognize semantics simultaneously.Meanwhile,speech recognition merely analyzed semantics,but failed to recognize the identity simultaneously.In this thesis,firstly,technology development in voiceprint recognition has been reviewed in the aspects of voice collection,pre-process,voiceprint feature extraction,identification and match.We propose an automatic voiceprint recognition method based on waveform analysis,where voiceprint and semantics can be recognized simultaneously.The key is to employ wavelet decomposition to analyze unstationary signals i.e.voice messages,which are decomposed in time-frequency domain to extract the feature of semantics and voiceprint.And then we develop a new method by combining the process of lowering noises in voice and extracting feature of voiceprint at the same time in wavelet domain.This method reduces consumption of resources and makes voiceprint recognition more efficiently.By using this method and combining it with deep neural network in voiceprint recognition or speech recognition,or both of these two tasks,we can achieve a recognition rate of more than 95%.