Research And Implementation Of Speaker Recognition System Based On Embedded Platform

Speaker recognition system collects speech signal through the voice equipment, and then the speech signal is processed and extracted by the feature parameters, and finally, the identity of the speaker is identified by the algorithm of speaker recognition. As a hot research topic of today, speaker recognition has a broad application prospect, especially in areas such as justice, financial, and information services.In view of the current speaker recognition algorithm complex degree high, weak robustness, this paper mainly studies the speaker recognition algorithm, and from a practical point of view of, the design and implementation of the speaker recognition system based on Embedded Linux.This topic chooses Samsung S5PV210 as the core processor, and builds the embedded platform based on the embedded linux system. Use u-boot to develop bootloader bootstrap program, the kernel of linux-3.0.8 cut and compile to generate the kernel image file uImage, using NFS might root file system, finally the application mainly uses QT to develop the final interface of speaker recognition.In speaker recognition system, the subject uses the function library of ALSA to realize the collection of voice, to realize the pretreatment of speech signal and extract the characteristic parameters of the processed speech signal, finally using the principle of Vector Quantization (VQ) and Gaussian Mixture Model (GMM) to realize speaker recognition system, and through the other speaker recognition system such as Hidden Markov Model(HMM) and Artificial Neural Networks(ANN) to deepen the understanding of speaker recognition. At the same time, this paper uses spectral-temporal receptive fields(STRF), this algorithm is more in line with the ear characteristics than the Mel frequency coefficient of frequency (MFCC), which can improve the robustness of the system, but because of its low recognition rate, so it can be integrated with MFCC, so as to improve the recognition rate of the system. At the same time in order to improve the GMM model, the improved genetic algorithm and improved EM algorithm to use, finally the improved STRF use in improved GMM model, the proposed algorithm is implemented to.Through a series of transplant operation, the speaker recognition system implemented on PC is transplanted to the development board to realize its real application. The speaker recognition system based on embedded system has the advantages of real-time, specificity and good user interface, which lays a practical foundation for the use of the system in the future market.