Font Size: a A A

Voiceprint Recognition In Acoustic Control System Research And Implementation

Posted on:2013-02-03Degree:MasterType:Thesis
Country:ChinaCandidate:H N FuFull Text:PDF
GTID:2248330374954325Subject:Computer system architecture
Abstract/Summary:PDF Full Text Request
With the development of digital life and multimedia technology, looking for a morenatural, friendly, stable interaction mode, people are more and more eager to break thetraditional mode of man-machine interaction. Recently, iphone4s causes a new upsurgein studying speech recognition. This paper builds a voice control system, which is basedon Windows as the environment of the specific application of voiceprint recognition.Voiceprint recognition and speech recognition are joined up to achieve voice controlsystem with the function of voiceprint recognition and a mode of multi-user withdifferent operation grades. This paper focuses on the voiceprint recognition relatedtechnology research,in order to seek a humanized interaction mode that is based on thefunction of voiceprint recognition. The concrete content is as follows:Firstly, this paper introduces the theoretical underpinnings of voiceprintrecognition technology based on Gaussian Mixture Model (GMM) and voice controlbased on Speech SDK. In view of the weakness that traditional GMM need a largenumber of samples and GMM-UBM make speaker forced to obey the uniformdistribution. This paper proposes the modeling and recognition method ofdiscriminative GMM, which increases the differences among the speakers, so that spacedistribution of speaker’s feature vectors can be better fitted by the model. At the sametime, the two sub models of UBM produces in the modeling process are the classifier offeature space and gender, so as to improve the recognition rate and speed of response.Then in view of the weakness that traditional K-means clustering algorithm onlyhas the ability of the local optimization and is sensitive to the initial centers and noisepoint. Two improved clustering algorithms applying for voiceprint recognition areproposed. They are clustering algorithms based on density and weighted distance ofvariance and a global optimize approach based on simulated annealing. Through theclustering algorithm is presented in this paper, Elliptic distribution of high dimension ofspeech data forms more accurately acoustic feature classes, therefore each Gaussiantakes more accurate data in order to improve the performance of the system. Meanwhile, the algorithm can also be extended to other areas.Finally, this paper build a voice control system, which have the function ofvoiceprint recognition is based on Windows API and message response mechanism.More important point is that multi-user with different operation grades is designed insystem. Completing the goal of man-machine interaction based on the function of thevoiceprint recognition.
Keywords/Search Tags:Voice Control, Voiceprint Recognition, GMM-UBM, DiscriminativeGMM, Feature classes
PDF Full Text Request
Related items