Research And Implementation Of The Voiceprint Recognition System Based On Android

With the rapid development of smart phones and mobile internet,traditional passwords can not meet the user's demand for the reliability and convenience of information security.Digital biometric authentication,which is represented by voiceprint recognition,has become a research hotspot in the field of artificial intelligence.With the rapid popularity of Android system,voiceprint recognition systems based on mobile terminals have a broad application prospects in digital security authentication.However,voiceprint recognition technology is facing with problems of poor real-time performance and robustness in the Android platform with limited system resources.In this paper,the research mainly concentrates in the following respects:The speech signal feature extraction algorithm is studied.For the traditional Mel Frequency Cepstrum Coefficients(MFCC)feature extraction algorithm on the Android platform,the computing time is longer and the robustness in the noise environment is poorer.This paper combines its first-order and second-order differentials to obtain an improved combination of coefficients.Combined the advantages of Android platform,this paper proposes a multithreading concurrency algorithm for feature extraction.The feature extraction task is decomposed into n equal parts and the corresponding n sub-threads are started for concurrent calculations.Applying the improved feature coefficients to the Android platform,the equal error rate is reduced by 7.24%and the recognition rate is increased by 6.8%,with a significant improvement of robustness in noise environment.After adding three sub-threads,the feature extraction calculation time is shortened by about 35%.The speech feature vector pattern matching algorithm is studied.Since the traditional modeling and matching algorithm is not suitable for the local real-time application of the speaker identification system in the mobile platform,this paper proposes a dynamic threshold dual-layer pattern matching algorithm based on Dynamic Time Warping(DTW)and Gaussian Mixture Model(GMM).In the first layer,the DTW algorithm is used to rapidly screen the model library to narrow the matching range,and the second layer was accurately matched by the GMM to make the final judgment.The algorithm can reduce the calculation time by about 41%-55%without excessive loss of recognition accuracy compared with the traditional GMM matching method,and the accuracy is increased by about 12%to 14%compared with the traditional DTW matching algorithm.According to the number of the registration model,the first layer thresholds is dynamically adjusted to obtain the highest accuracy in the shortest time,realizing the real-time application of mobile terminals.An Android voiceprint recognition system applied in the field of industrial automation is designed and implemented.Aiming at the low degree of automation of Android device audio function testing in the current production line,this paper proposes an audio fault diagnosis system based on voiceprint recognition technology for Android devices.This system uses MFCC and its first-order differencial as the features coefficients and the DTW algorithm is used to compares the model features of voice recorded by microphone and broadcasted by loudspeaker,so as to effectively detect the various malfunctions of the speaker and microphone simultaneously including soundlessness,distortion,noise and imbalance.The detection accuracy of the system ranges from 100%to 93.33%,which is 6.25 times higher than the efficiency of manual tests,bringing the company for tangible benefits over 90kUSD/year according to the preliminary estimates.Through lateral comparison with other literatures,it is proved that this system has significant advantages in fault diagnosis type and low cost.