标题: Research on Voiceprint Recognition System and Pattern Recognition Algorithm[查看完整版帖子][打印本页]
时间: 2012-9-27 15:52
作者: qiu810 标题: Research on Voiceprint Recognition System and Pattern Recognition Algorithm
Research on Voiceprint Recognition System and Pattern Recognition Algorithm
ABSTRACT
The identity recognition technology based on biometric characteristics was an important issue in the recent international research. Voiceprint recognition, in which the identity of the speaker was determined by voice recognition, had the widespread application value in the fields of System Security Authentication, Judicial Identification, Electronic Interception，etc.
Voiceprint recognition which was a type of voice recognition can be classified into two categories, speaker verification and speaker identification from the view of application as well as recognition with relevant test and recogntion with irrelevant from the view of recognition condition. In voiceprint recognition individual information features were focused, ignoring the content of the voice signal. There were two key technologies in voiceprint which were feature extraction in which the speaker’s voice characteristics were described by the feature parameters extracted from the voice signal in acoustic or statistical terms, as well as recognition model by which robot could learn and memorize the speaker’s characteristics in order to realize the recognition of the speaker.
This paper demonstrated the principles of voiceprint recognition technology and emphasizes on studying as follows:
(1) Extraction of voice signal: keynote period, zero-crossing rate, brightness, Linear Prediction Coefficients, LPC, Linear Cepstral Prediction Coefficients, LPCC, Mel-Frequency Cepstrum Coefficients, MFCC, etc
(2) voiceprint recognition approaches and models: Gaussian Mixture Models, Implicit Markov Model, Vectorization Model, Artificial Neural Network(ANN) Model, Support Vector Machine Model.
The recognition effect of the existing algorithms is susceptible to environmental noise, voice variation and other factors. According to this problem, Basing on the existing voice pattern recognition technology, this paper improved the calculation method and did plenty of experiments (experiments use the voice data which is gathered under different noise environment of the early, middle and late periods of a day. And during the one-month voice collection, the person got bad cold caused voice variation.). Experiment results showed that the improved algorithm can effectively overcome the impact of environment noises and voice variation. This paper did work as follows:
(1) MFCC extracted form voice characteristics was improved, while the impact on voice signal was reduced by applying frequency masking algorithm; The accuracy of calculation was resolved in LF, HF, MF respectively, in turn the recognition rate was improved to a certain extent.
(2) A novel method of Initial codebook selection was presented by improving Vector Quantization Model: With the Hypersphere Extreme Selection Method and the improved LBG algorithm, the produce of empty cell during the converging process was reduced with an effectively improved recognition rate.
(3) Applying Labview graphical programming into the system of voiceprint recognition, a graphical virtual instrument panel was established by using powerful graphical environment and hardware resources in order to realize the real time selection and analysis of voice signal and the modularization, intellectualization through other softwares at the advantage of low cost, convenient analysis of statistic,good management.
KEY WORDS: Voiceprint Recognition; Mel-Frequency Cepstrum Coefficients; Vector Quantifying Model ; LabVIEW