Research Of Voiceprint Recognition Algorithm In Urban Traffic Environment

Voiceprint recognition technology uses personality factors contained in thespeech signal to recognize the type of target. In the quiet laboratory environment, ithas achieved satisfactory result. However, robust voiceprint recognition technologyunder noisy environment is still a pressing problem, with the urgent requirement ofpractical voiceprint recognition technology.With the rapid development of intelligent vehicles and unmanned vehiclestechnology, the voiceprint recognition technology also attract wide attention as animportant part of them. Exploring healthy voiceprint recognition technology, that canbe used on the intelligent vehicles or unmanned vehicles has become a research hotspot. Therefore, the noise complex and variety under urban traffic environment bring agreat challenge to voiceprint recognition system.Robust front-end processing algorithms of voiceprint recognition under urbantraffic environment are proposed, based on the human auditory system. It caneffectively reduce the impact of noise. The main work and innovation are as follows.(1) L-CASA speech separation algorithm is proposed based on the Hu-Wangmodel and the membership function. Simulation results show that the algorithm iseffective and superior t in the noise conditions.Analysis and evaluate computational auditory scene analysis speech separationalgorithm deeply. Then build a new hearing oscillation model based on membershipfunction. Combined with Hu-Wang model, L-CASA algorithm is proposed as a newcomputing auditory scene analysis algorithm. It effectively improves the anti-jammingcapability of voiceprint recognition system under urban traffic environment in whichvarious types of noise (white, street, car noise) exist.(2) GFCC-REC parameter and GF-MAR parameter, contained more voiceinformation, are proposed on the base of auditory peripheral model and reconstructionalgorithms. Simulations in noisy conditions verify the effectiveness and robustness ofthe parameters.The auditory peripheral model built in this paper is on the base of Gammatonefilter and Meddis model. With the auditory peripheral model, a robust auditory feature parameter, GF parameter is extracted. Furthermore, use the reconstruction algorithmto extract GFCC-REC parameter and use the boundary smoothing algorithm to extractGF-MAR parameter. Both the two parameters can effectively reduce the informationloss due to noise interference. Simulation results show that GFCC-REC parameter andGF-MAR parameter have better recognition performance in different noiseenviroments, compared with GFCC parameter and MFCC parameter. GF-MARparameter and the GFCC-REC parameter have the advantages of high robustness andlow complexity.(3) Build a voiceprint recognition system by the use of the front-end processingalgorithm, containing L-CASA separation and the GFCC-REC, GF-MAR featureextraction algorithm. The voiceprint recognition system is highly robust in urbantraffic environment.Use the front-end processing algorithm, containing L-CASA separation and theGFCC-REC, GF-MAR feature extraction algorithm, in the voiceprint recognitionsystem. It is worth to emphasize that the two parameters (GF-MAR, GFCC-REC) areeffectively integrated in the proposed voiceprint recognition system on the base ofscore fusion. The simulation results show that the proposed voiceprint recognitionalgorithm has high robustness and performance in urban traffic noise environment.The research improves the robustness of the voiceprint recognition technology inthe complex noise environment, and also promotes the practical development ofvoiceprint recognition in urban traffic environment. It provides new technical supportfor the using of voiceprint recognition on intelligent vehicles and driverless cars.