The Design Of Voiceprint Feature Extraction SOPC Based On Muilti-parameter Improved MFCC

Voiceprint authentication technology which contains biometric of speaker and only needs to extract the speaker's voice when in use has great potential in the future "Internet + " applications with the ability to take advantage of low cost, convenient compared with other authentication such as face, fingerprint, iris, etc. Voiceprint feature extraction is one of the key technologies to voiceprint recognition system, SOPC provides a new platform for speaker voiceprint feature extraction application research, voiceprint feature extraction IP core is integrated to complete high-performance voiceprint authentication in portable devices has become the hot topic.To meet the voiceprint feature extraction of embedded system requirements, multi-parameter improved MFCC voiceprint feature extraction SOPC design was proposed on the basis of in-depth study of voiceprint feature extraction, multi-parameter mainly refers to the pitch in the time domain, spectral amplitude in the frequency domain and FFT points. In this paper simulation algorithms, hardware design, IP core package, system testing and other related content were studied:1. After in-depth study of traditional Mel filter banks, in order to improve the adaptability of Mel filter banks for high accuracy, improved Mel filter banks based on multi-parameter was designed for the problem of speaker verification lower precision caused by fixed Mel frequency band of traditional Mel filter banks. Pitch extracted in time Domain was to improve frequency bandwidth of Mel filter banks, the amplitude spectrum and FFT transform points extracted in the frequency domain were used to make the output of Mel filter banks better. Pitch of speaker speech was extracted by the low-pass filter, the amplitude spectrum and FFT points were got by the FFT transform to the pre-processing voice data.2. The completion of Mel filter banks package was finished by Nios?soft core and IP core technology after the completion of modules design with the top-down method. Combined with Log, DCT hardware modules, the integral voiceprint feature extraction SOPC system was build to realize the functions of the voice data input and features output.Experiment results show that the speaker verification accuracy of the use of MFCC extracted with multi-parameter improved Mel filter banks is higher than the use of traditional Mel filter banks, multi-parameter improved MFCC voiceprint feature extraction SOPC design in this paper provides a theoretical basis and a new case study for integration design of feature extraction and realization of voiceprint authentication on the internet.